SynthUSBII USB RF Signal Generator & CPDETLS-4000 RF Power Detector

[TomC]

My old Heathkit RF SG tops at 110 MHz and for a while I've been wanting to upgrade so I can experiment with higher frequencies, but my wife would have my head if I invested on a top of the line HP for my hobby! Even an older used model from eBay! 

So when I saw this Windfreak SynthUSBII RF SG at Saelig for $199 during Cyber week I decided to take a chance and see what it could do!

Attachment #1: This is a picture of the SynthUSBII obtained from the link below.

https://windfreaktech.com/product/usb-rf-signal-generator/

The unit is very small, about 5 x 2.5 cm and is controlled by the included software. One caveat is that although you can set the power output to one of four levels, there is no way to know with reasonable accuracy what those levels are.

Attachment #2: According to the manufacturer the SynthUSBII frequency response should be similar to this. I would like to verify that and see how close my unit is to the published response curve. I suspect it will vary from unit to unit, but I'll post whatever results I come up with for my unit. I would also like to know what the response curve for the other power levels looks like.

Not that knowing the output power at the SMA connector would be that helpful in every case, since at the higher frequencies the power arriving at the DUT may be a different level due to losses loading etc.! So it would be better, in my opinion, to measure the power level as close to the DUT as possible.

After playing with the unit for a while I decided I also needed a way to check the power output. My OWON SDS7102 DSO can do a reasonable job to around 100MHz, but beyond that readings to the maximum usable frequency of about 350 MHz are of uncertain accuracy, and what about the frequencies to the SynthUSBII limit of 4.4 GHz. Using an RF probe didn't seem to be a reasonable alternative, I own a couple of those and they did a dismal job! And buying a calibrated power meter for that frequency range will get me back to that very uncomfortable head chopping situation!

So when I saw the CPDETLS-4000 RF Power Detector at Digi-Key for $31 I decided to give it a try!

http://www.digikey.com/product-search/en?WT.z_cid=sp_744_0310_buynow&site=us&lang=en&mpart=CPDETLS-4000

Attachment #3: This is a picture of the CPDETLS-4000 obtained from the manufacturer's website.

This device covers a frequency range of 10MHz to 4 GHz, so it seemed to me that it would be a perfect, inexpensive companion, for the SynthUSBII. One caveat is that the manufacturer's datasheet doesn't mention the accuracy or tolerance of the given tabulated values. However, since they used 3 decimal places I'm going to embrace a little wishful thinking and hope that they are extremely accurate!

Attachment #4: These are the tabulated values for Frequency & Power in versus DC output that appear in the datasheet

As I experiment with these devices I plan to post my findings in this thread.

[saelig]

My old Heathkit RF SG tops at 110 MHz and for a while I've been wanting to upgrade so I can experiment with higher frequencies, but my wife would have my head if I invested on a top of the line HP for my hobby! Even an older used model from eBay! 

So when I saw this Windfreak SynthUSBII RF SG at Saelig for $199 during Cyber week I decided to take a chance and see what it could do!

Attachment #1: This is a picture of the SynthUSBII obtained from the link below.

https://windfreaktech.com/product/usb-rf-signal-generator/

The unit is very small, about 5 x 2.5 cm and is controlled by the included software. One caveat is that although you can set the power output to one of four levels, there is no way to know with reasonable accuracy what those levels are.

Attachment #2: According to the manufacturer the SynthUSBII frequency response should be similar to this. I would like to verify that and see how close my unit is to the published response curve. I suspect it will vary from unit to unit, but I'll post whatever results I come up with for my unit. I would also like to know what the response curve for the other power levels looks like.

Not that knowing the output power at the SMA connector would be that helpful in every case, since at the higher frequencies the power arriving at the DUT may be a different level due to losses loading etc.! So it would be better, in my opinion, to measure the power level as close to the DUT as possible.

After playing with the unit for a while I decided I also needed a way to check the power output. My OWON SDS7102 DSO can do a reasonable job to around 100MHz, but beyond that readings to the maximum usable frequency of about 350 MHz are of uncertain accuracy, and what about the frequencies to the SynthUSBII limit of 4.4 GHz. Using an RF probe didn't seem to be a reasonable alternative, I own a couple of those and they did a dismal job! And buying a calibrated power meter for that frequency range will get me back to that very uncomfortable head chopping situation!

So when I saw the CPDETLS-4000 RF Power Detector at Digi-Key for $31 I decided to give it a try!

http://www.digikey.com/product-search/en?WT.z_cid=sp_744_0310_buynow&site=us&lang=en&mpart=CPDETLS-4000

Attachment #3: This is a picture of the CPDETLS-4000 obtained from the manufacturer's website.

This device covers a frequency range of 10MHz to 4 GHz, so it seemed to me that it would be a perfect, inexpensive companion, for the SynthUSBII. One caveat is that the manufacturer's datasheet doesn't mention the accuracy or tolerance of the given tabulated values. However, since they used 3 decimal places I'm going to embrace a little wishful thinking and hope that they are extremely accurate!

Attachment #4: These are the tabulated values for Frequency & Power in versus DC output that appear in the datasheet

As I experiment with these devices I plan to post my findings in this thread.

Hi There!

Thanks for buying the amazing SynthUSBII (http://www.saelig.com/windfreak/synthusbII.htm) in our Saelig CyberWeek Sale!  I contacted David Goins (Windfreak Technology CTO) and he makes this comment:  "Those detectors are sensitive to output loading.  In my experience they are not just sensitive to the load resistance, but can also vary just based on the length of the output cables that are measuring that voltage.  It would be nice if the datasheet mentioned what measurement device was used – was it a 1Mohm input impedance volt meter??

That’s been my experience using them, but maybe this one is better.  But there is no active device / buffer to factor out the RF impedance of the diode inside that device from the load impedance, which makes the system a voltage divider…  If you could get the exact setup from the manufacturer then you'll probably get accurate enough results..""

Hope that helps!  I'll be interested to see your results posted here!

Alan Lowne  CEO Saelig Co. Inc.

[TomC]

Hi There!

Thanks for buying the amazing SynthUSBII (http://www.saelig.com/windfreak/synthusbII.htm) in our Saelig CyberWeek Sale!  I contacted David Goins (Windfreak Technology CTO) and he makes this comment:  "Those detectors are sensitive to output loading.  In my experience they are not just sensitive to the load resistance, but can also vary just based on the length of the output cables that are measuring that voltage.  It would be nice if the datasheet mentioned what measurement device was used – was it a 1Mohm input impedance volt meter??

That’s been my experience using them, but maybe this one is better.  But there is no active device / buffer to factor out the RF impedance of the diode inside that device from the load impedance, which makes the system a voltage divider…  If you could get the exact setup from the manufacturer then you'll probably get accurate enough results..""

Hope that helps!  I'll be interested to see your results posted here!

Alan Lowne  CEO Saelig Co. Inc.

Hi Alan,

Thanks for the info!
Sorry I didn't respond sooner, I got sick with the flu and just started feeling better!

My preliminary testing indicates that this detector also exhibits some sensitivity to output loading, so far it seems to be limited to about a 2% change. I've been using a DMM with a 2.5G ohm input impedance, but haven't decided yet if I want to stick with it or use a different instrument for the final measurements.

Anyway, I plan to post pictures of the final setup. Also, I'm working on a spreadsheet based on the datasheet's tabulated values that uses interpolation to provide values in between the given ones (see attachment #1).

As far as the detector's input interface, I believe, since the manufacturer specifies it as 50 ohm, that it must contain some type of input matching network (see attachment #2). However, when directly connected to the output of the SynthUSBII, my preliminary testing indicates that it doesn't quite behave the same as say, for example, a resistive 50 ohm terminator.

By the way, I also bought an Owon AG1012F during Saelig's 2014 CyberWeek sale! In case you are interested, here is a link to the thread where I describe my experience and document many tests. Warning, this is not as rosy as the reviews posted on your site!

https://www.eevblog.com/forum/testgear/owon-ag1012f-arbitray-waveform-generator/

[TomC]

On this post I'd like to explore the possibility of using the SynthUSBII & CPDETLS-4000 to get a better idea of the frequency response of a DSO beyond the manufacturer's specified bandwidth. Hopefully, this knowledge can help the user determine the amplitude of signals that exceed the DSO's specified bandwidth and thus extend the usefulness of inexpensive lower bandwidth DSOs. As an example I'll be using my Owon SDS7102 DSO for the following tests.

For this experiment as well as for future experiments I'm relying on the CPDETLS-4000 datasheet's tabulated values (see the first post). To get values in between the published values I created a spreadsheet that uses linear interpolation. Frequency is tabulated at 5MHz intervals from 10MHz to 1GHz and at 10MHz intervals from 1GHz to 4GHz. Power is tabulated at 0.01 dBm intervals from -10 dBm to 10 dBm and there is an extra column with the corresponding mVpp values. The spreadsheet is in Open Office .ods format and is too large to attach to this post, so I uploaded the two versions to OneDrive for anyone that may want a copy:

CPDETLS-4000_new.ods - This is the full version that includes the interpolation formulas.

https://onedrive.live.com/redir?resid=967A90CA47FD025B!204&authkey=!AB5PBzQdKrjbsuM&ithint=file%2ctxt

CPDETLS-4000 compact plus.ods - This version has the same functionality but instead of the formulas it only has the values. The advantage is that it's more compact and loads faster.

https://onedrive.live.com/redir?resid=967A90CA47FD025B!203&authkey=!ALlQueFdX8yHsGY&ithint=file%2ctxt


Attachment #1 - This shows how the SynthUSBII & CPDETLS-4000 were connected to the DSO for this experiment. After trying a number of different adapters and cables it became evident that the implementation of this connection has a significant impact on the readings. I opted for the adapters that provided the shortest possible signal path between the elements. In my view this setup resulted in more credible results in that there was a closer match between the DSO's and CPDETLS-4000's readings in the 35MHz to 100MHz frequency range. Although, in my opinion, connecting these 3 devices together without a splitter results in an impedance mismatch. I believe that keeping the signal path as short as possible minimizes the impact of signal reflections on the readings.


Attachment #2 - This shows how the DMM & CPDETLS-4000 were connected for this experiment. The implementation of this connection didn't seem to be as critical although during earlier experiments I though I had detected some output loading sensitivity. As it turns out, I now believe that what I saw was due to slight differences in output power as the SyntUSBII warms up. Later observations confirmed that the SynthUSBII output power slightly increases as it warms up, however, after about one hour it seems to become nearly stable. The DMM reading shown on this picture (47.67mV) was taken during the 250MHz test. The reading jumps around slightly, so for more stable readings I used the DMM's MAXMIN feature to pick the Maximum value.


Attachment #3 - This shows the SynthUSBII PC software setup for the 250MHz test. For this experiment I started with a 35MHz setting and then increased the frequency in 5MHz steps until reaching 350MHz. I chose this limit because the SDS7102 DSO seems to be able to display frequencies of up to around 350MHz without noticeable evidence of aliasing.


Attachment #4 - This shows the SDS7102 DSO's screen during the 250MHz test. The SynthUSBII output tends to jump around slightly, so I decided to set the DSO to Average 16 for more stable readings. If the Vp measurement (Vp means Vpp in Owon lingo) still varied I picked the lowest reading. In this case the DSO reports that this signal is only 302mVpp, but as will see the CPDETLS-4000 reports a higher value.


Attachment #5 - This shows the area of the spreadsheet that allowed me to determine the dBm and mVpp values reported by the CPDETLS-4000 during the 250MHz test. I picked the value closest to the DMM's reading, in this case, since the DMM read 47.67mV I picked the 47.656 value on the 250MHz column. So in this case the CPDETLS-4000 reports a 485.9mVpp signal (-2.29dBm) as opposed to the 302mVpp reported by the DSO. From these figures we can calculate a % Error (percentage error) that could be used later to deduce the amplitude of 250MHz signals observed with the DSO. In this case the % Error is:

      (485.9 - 302)/302 x 100 = 60.894%

What this means is that the amplitude of 250MHz signal readings obtained with the DSO is actually 60.894% higher than what the DSO reports. For example, if the DSO reading is 302mVpp then the actual signal amplitude is:

      302 x 0.60894 + 302 = 485.9mVpp

And if the DSO's reading is 400mVpp then the actual signal amplitude is:

      400 x 0.60894 + 400 = 643.576mVpp

The spreadsheet includes a facility that allows the user to calculate % Error or signal amplitude by just entering the known values.


Attachment #6 - This is a line chart of the SDS7102 DSO (blue) and CPDETLS-4000 (red) readings I obtained during this experiment. The line chart also includes the % Error for each pair of readings (green). Note that the Y scale for the DSO & CPDETLS-4000 readings is on the left and the Y scale for the % Error is on the right.

Looking at the % Error curve we can see that up to about 200MHz the readings reported by the SDS7102 DSO & the CPDETLS-4000 are within plus or minus 10% of each other. In my opinion this is probably within the tolerance range of the readings considering the way the test was implemented. So I wouldn't know which reading from each pair should be picked as more accurate.

However, from 200MHz to 350MHz it seems evident to me that the DSO's response is on a fairly steep roll-off and the CPDETLS-4000 readings are the most accurate within this range. So I think that in this range it would be appropriate to use the % Error and the associated SDS7102 DSO reading to deduce values closer to the actual signal amplitude. Still, keep in mind that there is a tolerance associated with the deduced amplitude, probably in the range of plus or minus 10%.


Attachment #7 - This is the spreadsheet used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during this experiment.


Some background on why sometimes it's possible to extend the usefulness of DSOs (the way I see it)
--------------------------------------------------------------------------------------------------

Like most modern DSO's, the SDS7102 uses sin(x)/x interpolation to reconstruct the input signal. Since the single channel sampling rate is up to 1GS/s, the Nyquist frequency is 1G/2 = 500MHz. So in theory, provided a rectangular filter that totally rejects all frequencies above 500MHz is used, it should be able to faithfully reconstruct input frequencies of up to 500MHz.

Unfortunately close to perfect filters are expensive and difficult to attain, so to cut cost manufacturers may rely on a number of compromises to get the most cost effective results. For example, instead of a rectangle the filter may roll-off slowly and as the frequency of the input signal increases the signal amplitude displayed by the DSO will be proportionally lower. In addition, the filter may not be able to totally reject everything above the Nyquist frequency. As a result, as the input frequency approaches the Nyquist frequency some higher frequency components will also make it through and cause aliasing.

To compensate for these shortcomings the manufacturer may use a sampling rate that exceeds the Nyquist criteria by a wide margin and specify the DSO's bandwidth well below the Nyquist frequency. Consider for example the SDS7102, the specified bandwidth is 100MHz but the Nyquist frequency is 500MHz. So at the bandwidth's highest frequency (100MHz) we get 10 samples per period instead of the Nyquist minimum requirement of 2 samples per period.

Would it be advantageous to know the filter's response curve?
-------------------------------------------------------------

I think so, specially on DSO's like the SDS7102 where the Nyquist frequency is much greater than the specified bandwidth. This may mean that the DSO's front end will pass frequencies above the specified bandwidth. Although at some point the filter's roll-off will cause attenuation distortion, if the user knows the details of this response curve it should be possible to deduce the amplitude of the displayed signal. This may not be perfectly accurate but still a lot closer to reality than what's displayed by the DSO.



   ---------------------------------------------------------------



Next I plan to repeat this experiment using the X10 scope probe

[TomC]

This post is basically a repeat of the previous experiment. The one notable difference is that instead of connecting directly to the DSO's BNC I'm using one of the stock X1 / X10 probes that came with the DSO. My SDS7102 came with Owon T5100 probes. According to the manual the system bandwidth when these probes are used is 6MHz on X1 and 100MHz on X10. For this experiment the probe will be set to X10. As in the previous experiment, amplitude values for input frequencies from 35MHz to 350MHz in 5MHz steps will be tabulated.


Attachment #1 - This shows how the DMM, SynthUSBII, CPDETLS-4000, and the T5100 probe were connected for this experiment. The implementation of this connection doesn't have nearly as much impact on the readings as the configuration used in the previous experiment. I believe this is due to the probe's 10 M ohm input resistance. In my opinion this higher value results in a less severe impedance mismatch when the 3 devices are connected together. However, I still opted for the adapters that provided the shortest possible signal path between the elements. In my opinion, as the frequency increases the probe's input capacitance (14.5pF - 17.5pF) will have a larger influence on the probe's input impedance resulting in a larger impedance mismatch. So, as in the previous experiment, I believe that keeping the signal path as short as possible minimizes the impact of signal reflections on the readings.


Attachment #2 - This is a line chart of the DSO & T5100 probe combination (blue) and CPDETLS-4000 (red) readings I obtained during this experiment. As in the previous experiment, the line chart also includes the % Error for each pair of readings (green). Note that the Y scale for the DSO & T5100 probe combination / CPDETLS-4000 readings is on the left and the Y scale for the % Error is on the right.

Looking at the % Error curve we can see that up to about 190MHz the readings reported by the DSO & T5100 probe combination and the CPDETLS-4000 are within plus or minus 10% of each other. In my opinion this is probably within the tolerance range of the readings considering the way the test was implemented. So I wouldn't know which reading from each pair should be picked as more accurate.

However, from 190MHz to 350MHz it seems evident to me that the DSO & T5100 probe combination's response is on a fairly steep roll-off and the CPDETLS-4000 readings are the most accurate within this range. So I think that in this range it would be appropriate to use the % Error and the associated DSO & T5100 probe combination reading to deduce values closer to the actual signal amplitude. Still, keep in mind that there is a tolerance associated with the deduced amplitude, probably in the range of plus or minus 10%.


Attachment #3 - This is the spreadsheet used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during this experiment.



   ---------------------------------------------------------------



Next I plan to repeat this experiment using other X10 scope probes that are rated by the manufacturers as 200MHz or more.

[TomC]

I added some features to the graphs and spreadsheets discussed on the previous two posts.

The new graphs contain a plot of the BNC's or probe tip's "Worst Case Input Impedance" as the frequency increases. Keep in mind that this is "worst case". Some manufacturers provide impedance plots for their probes that take into consideration other factors, not just the input capacitance. The plots given here are a rough estimate based on a simpler model. However, I hope they can help illustrate the surprising amount of loading, and therefore signal degradation, associated with either the DSO's BNC or most passive X10 probes when used to visualize high frequencies. The plot's Y scale (right hand side of the chart) is in ohms. Note that there are passive X10 probes with very low input capacitance, < 4pF (e.g. Tek TPP1000, TPP0500B, and TPP0250), but expect to pay premium prices for them.

I also added a plot of the difference between the signal amplitude seen by the DSO or DSO probe combination and the signal amplitude seen by the CPDETLS as the frequency increases. The plot's Y scale (right side of the chart) is in dBm. This could be used to estimate bandwidth as per the 3dB down industry standard. However, keep in mind the uncertainty factor associated with the measurements made in this experiment. There is also an orange shaded area that is intended to make it easier to identify the part of the dBm Difference plot that corresponds to the -3dBm to 3dBm range.

Note that a bandwidth specification doesn't mean that the instrument will faithfully reproduce any signal that falls within that frequency range. In essence, it only identifies the frequency at which the amplitude of a sine wave as seen by the instrument can be expected to be 3dBs below the sine wave's actual amplitude. This is equivalent to saying that the sine wave seen by the instrument is about 0.707 (1/2^0.5) of the sine wave's real amplitude. For example, the graph for the Owon T5100 probe shows a 3.01dBm difference at 225MHz and the plot for the CPDETLS shows that it detects 537.1mVpp at this point. If we assume that the CPDETLS is correct and the sine wave's actual amplitude is 537.1mVpp, then we can expect that the DSO probe combination will see an amplitude of 537.1 X 0.707 = 379.8mVpp. This is exactly the amplitude that the DSO + probe plot shows at this point.

Finally, I also added a green shaded area to the graph, see attachments #1 & #3. This is intended to make it easier to identify the part of the % Error plot that corresponds to the plus or minus 10% range. As I indicated in a previous post, in my opinion, due to the way the test was implemented, when looking at pairs of mVpp readings within this range picking one reading or the other as the most accurate is just a guess. So I regard this as an uncertainty factor when evaluating the results of this experiment.



Attachment #1 - DSO Response at the BNC. This is the replacement for the line chart identified as Attachment #6 back on Reply #3. The difference is that it now contains a plot of the BNC's Worst Case Impedance as the frequency increases and a plot of the difference between the DSO and CPDETLS amplitudes in dBm. Both plots use the Y scale on the right. As before, this scale is also used for the % Error plot. The Y scale on the right provides adequate resolution for the % Error and Worst Case Impedance plots, however, the range of the scale used by the dBm plot is small and as a result the plot lacks detail. To get around this I created a duplicate line graph, Attachment #2, where the Y scale on the right is expanded to better accommodate the dBm difference plot.


Attachment #2 - DSO Response at the BNC with dBm difference expanded Y scale. This graph does not contain the BNC's Worst Case Impedance & % Error plots since they wouldn't fit due to the expanded Y scale.


Attachments #3 & #4 - Owon T5100 + DSO Response at the probe tip & Owon T5100 + DSO Response at the probe tip with dBm difference expanded Y scale. These replace the line chart identified as Attachment #2 on the previous post. They contain the same additional features as described above.


Attachments #5 & #6 - These are the spreadsheets used to plot the line charts. In addition to the line charts they contains all the tabulated readings obtained during the experiment as well as the tabulated values & formulas for the added features.

[TomC]

This post is about experimentally using the SynthUSBII SG and the CPDETLS-4000 RF power detector to check the high frequency response characteristics of scope probes. These are not the optimal instruments for this task, but in my opinion, they can produce moderately accurate results on a low budget. The test subjects are a number of inexpensive oscilloscope probes with alleged bandwidths of 200MHz or more that I have acquired during the last 2 years.


-------------------------------------------------------------------------------------------------------------------


I've known for some time that the SDS7102 DSOs had some surplus bandwidth. So during the last couple of years I've been shopping for inexpensive higher bandwidth probes in an attempt to better exploit this extra bandwidth. However, until now, all I was able to do to evaluate the new probes is compare them to the stock probes that came with the DSO using an uncalibrated 310kHz - 110MHz RF generator as the signal source. The results were at times frustrating and bewildering, since based on my comparisons the stock probes seemed to outperform the allegedly higher bandwidth probes. Unfortunately, due to my test equipment limitations, I couldn't even get an approximation of the actual bandwidth and response curve of both the new and the stock probes.

Now, with the help of the SynthUSBII SG and the CPDETLS-4000 RF power detector, I can get a better idea of the bandwidth and response curve of my new, allegedly higher bandwidth, probes. To test these probes I'll be using the same procedure I used for the stock Owon T5100 as discussed on the previous 2 posts. The probes will be very carefully compensated before performing each test. Compensation will be done with the probe connected to the 1KHz signal source via a probe tip adapter instead of using the probe clip and ground lead.


Li Hua P6200 200MHz passive probes bought on eBay
-------------------------------------------------

I bought these probes at eBay in June 2014 for $33.90 (free shipping). As soon as I tried them it became evident that these probes' bandwidth was nowhere near 200MHz. After contacting the seller I was issued an immediate refund and I was also allowed to keep the probes. The following is a link to the post where I describe the tests I performed on these probes at the time:

https://www.eevblog.com/forum/testgear/review-of-owon-sds7102/msg460398/?topicseen#msg460398

Now I have dug up these probes again to test them with my new gear.


Attachment #1 - Probe picture from the web.


Attachment #2 - Probe specs as shown in the manual. The specs given at the eBay page were identical.


Attachments #3 & 4 - Li Hua P6200 + DSO Response at the probe tip & Li Hua P6200 + DSO Response at the probe tip with dBm difference expanded Y scale.

Based on the response curve and dBm difference I think it's evident that these probes' bandwidth is way below 200MHz. I've seen a post by another member that tested the Li Hua 500MHz version (P6500) with equipment far superior than what I'm using here and reported results just as dismal.

https://www.eevblog.com/forum/testgear/oscilloscope-probes-41026/msg579768/#msg579768

The eBay seller from whom I bought these probes no longer sells this brand, however, if you search on Amazon for P6100, P6200, P6300, or P6500, you'll find plenty of marketplace sellers that offer them. Note that so far I haven't seen the brand Li Hua mentioned in the description, but look at the pictures closely and you'll see that they are the same as the ones pictured on #1.


Attachment #5 - This is the spreadsheets used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during the experiment.


250MHz passive probe bought from Saelig
---------------------------------------

I bought this probe from Saelig on December 2014 for $35. When I first tried it and compared it to my stock Owon T5100 probes I noticed that peak to peak readings were quite a bit higher in the 30MHz to 100MHz range with the largest difference at around 100MHz. However, due to test equipment limitations I couldn't confirm that these higher readings were less accurate than the ones obtained with the T5100 probes.

Now that I'm ready to test this probe with my new gear, I decided to investigate it further. First I found out that the probe pictured at the Saelig's website looks similar but is definitely not the same as the one I received. In addition, the specs given in the manual for the probe I received are slightly different than the specs given at the Saelig's website. For example, there is a 3pF difference on the input capacitance spec, also, the bandwidth spec is stated a little different. The manual identifies this probe as model 05SPGL250, however, the probe is labeled GLF-250. A web search revealed that this is a discontinued GW Instek probe intended for use with the GW Instek GOS-6200 scope (a discontinued 200MHz analog scope). In addition, the specs given on a Japanese GW Instek dealer's website once again are somewhat inconsistent with both the manual and the Saelig website.

The probe I received is nice looking and well built. However, in view of the above findings, I'm now inclined to believe that the exaggerated peak to peak readings are due to the fact that this probe was meant to be used with a different scope, namely, a GOS-6200. So I suspect that when used in conjunction with my SDS7102 DSO this probe's performance is subpar.


Attachment #6 - This shows a picture of the probe and its specs as it appeared at the Saelig website when I purchased it. Note that the input capacitance is specified as 11pF and the Bandwidth as 250MHz +/- 3dBs.


Attachment #7 - This shows a picture of the probe I received and its specs as it appeared at a Japanese GW Instek dealer's website. The Google translation is a little goofy (e.g. "start-up-time" instead of "rise time") but clear enough. Note that the GOS-6200 is the only applicable oscilloscope model mentioned. In addition, the specified input capacitance is different, 17pF instead of 11pF (my LC meter reads 16.2pF), and the bandwidth specification is stated differently, "DC - 250MHz" instead of "250MHz +/- 3dBs".

I now realize that when I purchased this probe I overlooked the +/- 3dB qualifier. Probably because I was expecting the bandwidth spec to adhere to the 3dB down industry standard. Inadvertently ignoring the added +/- 3dB amplitude distortion qualifier was a mistake I hope I won't forget. In the future I will be paying more attention to this red flag. I suspect that the +/- 3dB qualifier was added so that the probe could be marketed for use with scopes other than the GOS-6200. With this change the probe's frequency response is within specs even if the peaks and dips throughout the entire frequency range are as large as +/- 3dBs.

An industry standard bandwidth specification identifies the point where the signal amplitude versus frequency drops 3dBs below its low frequency value. A good quality probe is expected to exhibit a fairly flat response to at least 1/3 of its bandwidth followed by a slow roll-off to the 3dB down point. Another measure of quality is how well peaks and dips along the way are minimized. I don't think that +/- 3dBs peaks and dips would qualify as even mediocre quality.


Attachment #8 - Probe specs as shown in the manual. Note that the model number given is different from the label on the probe (GLF-250). Also note that the input capacitance is specified as 14pF (different from #6 & #7 above) and the bandwidth is stated slightly different, "DC to 250MHz +/-3dB".


Attachments #9 & #10 - GW Instek GLF-250 + DSO Response at the probe tip & GW Instek GLF-250 + DSO Response at the probe tip with dBm difference expanded Y scale.

Based on the response curve and dBm difference I think it's evident that this probe when used in conjunction with my SDS7102 overstates the amplitude of signals in the 50MHz to 120MHz frequency range. It also exhibits a sharper than expected roll-off starting at about 180MHz. Since the bandwidth spec on the probe's manual is given as "DC to 250MHz +/- 3dB" instead of the industry standard, the output signal amplitude may be up to 3dBs above or below the amplitude of the actual signal throughout the entire frequency range. Therefore, the frequency response curve is within specs since in this case, at its worst, it only overstates the 100MHz signal amplitude by a little less than 2dBm (about 120mVpp). I don't know what to say about this "within specs" performance, but the word crappy comes to mind! This makes this probe nearly useless when used in conjunction with my SDS7102. I suspect that when this probe is used in conjunction with other scopes, except the GOS-6200, the results will be similar.


Attachment #11 - This is the spreadsheets used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during the experiment.


Hantek PP-200 200MHz passive probes bought from Amazon
------------------------------------------------------

I bought these probes from Amazon.com in June 2014 for $31.99 (free shipping). When I first tried these probes they seemed to perform about on par with the Owon T5100 probes that came with my DSO. However, due to my test equipment limitations at the time, I couldn't evaluate their bandwidth beyond 110MHz. Now, with the help of my new gear, I hope to find out if their 200MHz rating is justified.


Attachment #12 - This shows a picture of the probes as they appeared at the Amazon.com page when I purchased them.


Attachment #13 - Probe specs as shown in the manual. The specs given on the Amazon.com page were identical.


Attachments #14 & #15 - Hantek PP-200 + DSO Response at the probe tip & Hantek PP-200 + DSO Response at the probe tip with dBm difference expanded Y scale.

Based on the response curve and dBm difference I think it's evident that, when used in conjunction with my SDS7102, the bandwidth of these probes is better than 200MHz. The response curve is similar to what I got with the Owon T5100 probes, but in some ways, in my opinion, is slightly better. For example, the dips and peaks along the 35MHz to 150MHz range are less pronounced. On the downside, the input capacitance spec for these probes is 18.5pF to 22.5pF compared to 14.5pF to 17.5pF for the T5100 probes. However, my LC meter reads 15.2pF for the PP-200 and 14.4pF for the T5100. So it appears that the actual input capacitance of the probes that I have is nearly the same.


Attachment #16 - This is the spreadsheets used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during the experiment.


------------------------------------------------------------------



Some things I learned along the way that in my opinion should be considered when trying to match new probes to an existing scope:

Matching a particular scope to probes from a different manufacturer can be a tricky endeavor. Although important, making sure that the compensation range matches the scope's input capacitance does not guarantee an adequate response curve. A probe by itself is not associated to a particular bandwidth, the specified bandwidth applies to the combination of the probe and a compatible scope. Claims that a probe is compatible with most scopes are probably exaggerated, specially if only LF compensation is available. Watch out for red flags, like an amplitude distortion qualifier added to the bandwidth spec (e.g. DC to 250MHz +/- 3dB). Search the web to see if someone has tested the response of the particular probe you are considering, other reviews based on non-technical observations can be deceiving.


Some things I'd like to try in the future:

The probes I tested so far only offer LF compensation. However, even if the probe's compensation range matches the scope's input, different scopes with the same input specs will still differ in their parasitic capacitance and inductance characteristics. As a result, unless the probe is specifically designed for a particular scope, the response curve obtained via LF compensation is a compromise and not the most optimal response curve possible. To get around this pitfall some probes offer HF compensation in addition to the standard LF compensation. The Pico Technology TA131 is an inexpensive example of a probe with this extra feature, so I'm currently considering purchasing one. This 250MHz probe with an 11pF input capacitance sells for about $42:

https://www.picotech.com/accessories/passive-oscilloscope-probes/250-mhz-scope-probe


And here is a link to an article that explains how to perform the HF compensation on a virtually identical probe:

https://www.picotech.com/library/application-note/how-to-tune-x10-oscilloscope-probes


---------------------------------------------------------------------


Next I'll try to plot the SynthUSBII response curve from 40MHz to 4GHz

[TomC]

This post contains the results of my attempt to plot the SynthUSBII response curve from 40MHz to 4GHz. The experiment was done with the SynthUSBII SMA-F RF-out connector directly attached to the CPDETLS-4000 SMA-M RF-in connector (see attachment #1). As can be seen the plots are not a close match to the response curve given by the manufacturer (see attachments #0 and #2). This, in my opinion, is not necessarily an indication that either one is inaccurate.

The published response curve, attachment #0, was probably obtained using a frequency independent load impedance, such as a 50 ohm resistor, and a very high impedance measuring instrument. In this case, since loading should be consistent, any signal amplitude changes throughout the frequency range can be attributed to SynthUSBII output signal variations. A frequency response curve describes the way the entire system reacts to frequency changes, in this case the system was designed so that amplitude changes are attributable to a single system element. Given these circumstances the resulting response curve can be called "SynthUSBII Output response Curve" or other similar name that indicates that the SynthUSBII is primarily responsible for any amplitude changes.

In contrast, although I don't know the exact implementation of the CPDETLS-4000 impedance matching circuit, I do know that the RF input measures about 6.5nF and > 20M ohms. So, given that reactive components are involved, I find it unlikely that its impedance matching circuit implementation can maintain a consistent load impedance of 50 ohms from 40MHz to 4GHz. As a result, it's my opinion that in this case the signal amplitude changes we see in the plots are not only attributable to SynthUSBII output signal variations but are also influenced by load impedance variations. I also get the impression that some of the peaks and dips may be attributable to parasitic resonances and reactances caused by circuit implementation geometries and component choices. Nevertheless, the resulting response curve still describes the way the entire system reacts to frequency changes. However, in this case, I chose to call the response curves "SynthUSB Output Power as Detected by a Directly Connected CPDETLS-4000 RF Power Detector". My intent was to explicitly identify the system used to obtain the curves since amplitude variations are not primarily attributable to a single system element.


Attachment #0: - This is the response curve given by the manufacturer.


Attachment #1: - This illustrates the way I connected the SynthUSBII and the CPDETLS-4000.


Attachment #2: - These are the response curves named "SynthUSB Output Power as Detected by a Directly Connected CPDETLS-4000 RF Power Detector". A separate response curve was plotted for each of the 4 SynthUSBII power settings. The orange curve corresponds to Power Setting 3, the red to 2, the brown to 1, and the black to 0. The gray plot on the background is the response curve from #0 that has been proportionally resized to fit the X and Y scales used on this line chart.


Attachment #3: - This is the spreadsheet used to plot the line chart. In addition to the line chart it contains all the tabulated readings obtained during the experiment.


                   -------------------------------------------------------------------


Plotting line charts by looking up DMM readings in the CPDETLS-4000 compact plus spreadsheet and manually entering the Vpp equivalent values is time consuming and at times boring! So I'm currently working on automating the process using an AutoIt script. The TENMA 72-7730A DMM I've been using has a USB interface and a logging software application that allows me to access the readings from my PC. So I'm hoping that the script can use this information, access the SynthUSBII application and read and/or set the frequency, and calculate the equivalent Vpp and dBm values. If there is any interest I'll post the code and executable once I have a working script!

[G0HZU]

The detector looks like it is a 50R load with a simple Schottky diode and it probably has a small amount of capacitance at the output.

The USB powered signal source looks like it is based on the ADF4351 chip and this will produce an output rich in harmonics. You could easily be looking at harmonics only 10-15dB below the main signal.

All of your plots appear to be using the detector as a linear detector because of the response curves provided with it.

So I think you are going to get confusing results when you connect the two devices together. The high harmonic content can produce high levels of uncertainty in the measurement (with this type of RF detector) and a lot depends on the phase relationship between the signals and also on how hard you drive the detector up in this linear region. The harmonics can make the detector read lower or higher than you might expect.

This all assumes that the RF detector is indeed a linear diode detector and (in theory at least) a linear diode detector (used as a power meter) can have approx +/- 20% of uncertainty if you have just one harmonic at -20dBc, or +/- 6% uncertainty if the harmonic is at -30dBc.

Because the ADF4351 output spectrum is so rich in harmonics I think you will have to filter the output before connecting the diode detector to it or the results you get for 'power' will have too much uncertainty to be of any practical use. You really need to use clean sine wave signals with this type of RF detector at the power levels you are using.

[G0HZU]

You can reduce the effect of all this by operating the RF detector at lower power levels such that it operates well below the linear mode. But you would ideally need to be feeding it with RF signals that are equal or smaller to the thermal voltage Vt and this is 25mV which is about -20dBm. The detector will then detect the influence of harmonics 'correctly'.

However, down at these power levels the detector will be very sensitive to temperature changes and this would make the detector fairly impractical unless you corrected for temperature somehow. You can often extend this low level (or 'square')detection range by playing with the load resistance at the output of the detector but probably only by 10-15dB.

[TomC]

Hi G0HZU,

Thanks for your insightful response, you are obviously well versed on RF technology! I'm just starting to get my feet wet on this fascinating field!

The detector looks like it is a 50R load with a simple Schottky diode and it probably has a small amount of capacitance at the output.

According to the manufacturer this detector contains a zero bias Schottky and a 100pF output capacitor (which they call video capacitance). The data sheet also identifies the RF input as 50 ohms, but there are no other details as to how this RF load is implemented. My LC meter reads about 6.5nF and my DMM reads open circuit in the 20M range. So if there is a 50 ohm resistor in there it must be connected in series with the capacitor. Maybe something like the circuit on attachment #1?

The USB powered signal source looks like it is based on the ADF4351 chip and this will produce an output rich in harmonics. You could easily be looking at harmonics only 10-15dB below the main signal.

All of your plots appear to be using the detector as a linear detector because of the response curves provided with it.

So I think you are going to get confusing results when you connect the two devices together. The high harmonic content can produce high levels of uncertainty in the measurement (with this type of RF detector) and a lot depends on the phase relationship between the signals and also on how hard you drive the detector up in this linear region. The harmonics can make the detector read lower or higher than you might expect.

This all assumes that the RF detector is indeed a linear diode detector and (in theory at least) a linear diode detector (used as a power meter) can have approx +/- 20% of uncertainty if you have just one harmonic at -20dBc, or +/- 6% uncertainty if the harmonic is at -30dBc.

You can reduce the effect of all this by operating the RF detector at lower power levels such that it operates well below the linear mode. But you would ideally need to be feeding it with RF signals that are equal or smaller to the thermal voltage Vt and this is 25mV which is about -20dBm. The detector will then detect the influence of harmonics 'correctly'.

However, down at these power levels the detector will be very sensitive to temperature changes and this would make the detector fairly impractical unless you corrected for temperature somehow. You can often extend this low level (or 'square')detection range by playing with the load resistance at the output of the detector but probably only by 10-15dB.

As you surmised the SynthUSBII is based on the ADF4351 chip, see attachment #2. As you know this chip produces square waves and in the SynthUSBII they are basically applied to the output unfiltered, so the odd harmonics are very prominent (See attachments #3 & #4 for an example at 100MHz).

Although I was already aware of the harmonics, I hadn't considered the fact that their effect could be different depending on the instrument used to measure the signal's amplitude. Before you brought this up I was thinking that since the harmonic content was present when the response curve given by the manufacturer was obtained, then, for comparison purposes, it shouldn't matter if it was present during my experiment.

As it turns out, the CPDETLS-4000 input to output ratio is frequency dependent. When a particular Vpp level is present at its RF input its DC output increases as the frequency increases for part of its frequency range, then it starts decreasing towards the end of the range. Depending on the harmonic content, as you predicted, this behavior will tend to overstate/understate the true signal level. With this new insight I now see that this behavior is at least partly responsible for the differences between the response curve given by the manufacturer and the response curves obtained during my experiment.

On the region of operation of the zero bias Schottky within the CPDETLS-4000 the only clue I have is the datasheet's Output Voltage vs Input Power table (see Attachment #5). Based on this I suspect that it operates on both the square law and linear portions as well as on the transition region in between (readers that may be wondering where these regions lie please see attachment #6). As I understand it, a detector's region of operation can be identified by its output voltage. For a typical zero bias Schottky detector 10mVs or less indicates that it's operating in the square law region, from 10mVs to 100mVs the transition region, and 100mVs or more the linear region (see attachment #7).

With this in mind it seems that I may be able to stay in the square law region throughout the entire frequency range by attenuating the SynthUSBII output so that it doesn't exceed about -8dBm (252mVpp). From #5 I can see that at this input power level the highest CPDETLS-4000 DC output is not far from 10mVs. So I plan to try to implement a test using this criterion to see how the shape of the response curve looks like. Perhaps, since the uncertainty introduced by the harmonics will be reduced as you explained, the response plot will be a closer match to the response curve given by the manufacturer.

Because the ADF4351 output spectrum is so rich in harmonics I think you will have to filter the output before connecting the diode detector to it or the results you get for 'power' will have too much uncertainty to be of any practical use. You really need to use clean sine wave signals with this type of RF detector at the power levels you are using.

Since the response curve supplied by the manufacturer includes the harmonic content, I don't think that filtering the harmonics would result in a closer match. However, for other applications, like checking the response of probes (previous posts) or filters, I totally agree that using clean sine waves would produce results with much less uncertainty. Now I need to start dreaming of a way to implement a tunable low pass filter that can be used to cover the entire frequency range!


Attachment #1 - This is a picture of an unrelated evaluation board. I'm using it to illustrate a 50 ohm impedance matching network, this particular implementation seems to consist of a microstrip a coupling capacitor and a resistor.


Attachment #2 - This is a picture of the SynthUSBII PCB.


Attachment #3 - Here the SynthUSBII is connected to the DSO's CH1 BNC via a 50 ohm feed-through terminator. The SynthUSBII is set to high power at 100MHz. The DSO is set to view the time domain.


Attachment #4 - Same setup as #3 but the DSO is set to FFT to view the frequency domain.


Attachment #5 - CPDETLS-4000 datasheet page showing the Output Voltage vs Input Power graph and table.


Attachment #6 - Illustration of a diode's VI curve where the square law and linear regions of operation are identified.


Attachment #7 - Illustration of a typical zero bias schottky diode's input power vs output voltage plot where the square law, transition, and linear regions are identified.

[G0HZU]

Since the response curve supplied by the manufacturer includes the harmonic content, I don't think that filtering the harmonics would result in a closer match. However, for other applications, like checking the response of probes (previous posts) or filters, I totally agree that using clean sine waves would produce results with much less uncertainty.

The problem is that the uncertainty from the harmonics is phase dependent and so a harmonic at -10 to -15dBc could cause quite a large window of uncertainty if you operate up in the linear region.

For example, if you were to deliberately change the phase of the -10dBc harmonic across a 360degree range then the detector voltage would change quite a bit. Possibly by >2dB in terms of indicated 'power' although it depends on how linear the detector is.

So even though the average power of the signals isn't changing in this test and 'only' the relative phase of the signals is changing, the detector will misreport a significant change in power as the phase is changed. As the phase could be fixed anywhere in this range on a 'real' test then you end up with huge amounts of uncertainty in the system if you don't know the phase relationship.

If you operate it down into the square law region things will get a whole lot better but then you may run into temperature issues because the detector response will shift wrt temperature quite a bit down in the square law region. You could introduce some passive compensation for this (and recalibrate) but this is still going to be a bit of a fudge.

[TomC]

The problem is that the uncertainty from the harmonics is phase dependent and so a harmonic at -10 to -15dBc could cause quite a large window of uncertainty if you operate up in the linear region.

For example, if you were to deliberately change the phase of the -10dBc harmonic across a 360degree range then the detector voltage would change quite a bit. Possibly by >2dB in terms of indicated 'power' although it depends on how linear the detector is.

So even though the average power of the signals isn't changing in this test and 'only' the relative phase of the signals is changing, the detector will misreport a significant change in power as the phase is changed. As the phase could be fixed anywhere in this range on a 'real' test then you end up with huge amounts of uncertainty in the system if you don't know the phase relationship.

If you operate it down into the square law region things will get a whole lot better but then you may run into temperature issues because the detector response will shift wrt temperature quite a bit down in the square law region. You could introduce some passive compensation for this (and recalibrate) but this is still going to be a bit of a fudge.

Thanks for your detailed explanation!

It inspired me to do some more research on the subject and to conduct some experiments. I plan to post the results as soon as I'm done.

[TomC]

On this post I describe a couple of recent experiments involving the SynthUSBII and CPDETLS-4000.

First I re-plotted the SynthUSBII response curve with a 4dB attenuator connected to it's output. The intent was to cause the CPDETLS-4000 to operate in the square law region. Admittedly, the attenuator I had on hand is not rated for GHz frequencies, but even in the low VHF region the results were dismal, a 3dB attenuator didn't do much better. The readings I obtained were erratic and quite a bit different than the levels reported by my DSO (see attachment #0).

Next I wanted to verify the accuracy of some of the calibration points given on the CPDETLS-4000 datasheet. The only signal source that I own that is suitable for this task is a 10MHz Dual Channel AWG, so I could only check the 10MHz calibration points. The results seem to indicate that the accuracy of the CPDETLS-4000 starts to degrade when the input power drops below 0dBm. By the time the input power drops to -10dBm the readings can be up to 25% off. Perhaps this has to do with drift caused by temperature changes when operating in the square law region as G0HZU suggested. However, the manufacturer claims an operating temperature range of -20 to 70 degrees Celsius and states that the instrument is suitable for "General Lab Use" (see attachment #1). So I think it would be reasonable to assume that some type of compensation was incorporated.

Another contributing factor may be the accuracy of my DMM, the manufacturer claims ±(0.05%+5) on the 200mV range, but I think readings of a few millivolts or fractions of a millivolt are probably not quite that accurate. Attachment #2 shows the results of this experiment.


Attachment #0 - SynthUSBII response curves. Orange, SynthUSBII set to power level 3 with a 3dB attenuator connected to its output. Red, SynthUSBII set to power level 3 with a 4dB attenuator connected to its output. I didn't check the full range of frequencies, it seemed pointless given the initial results.


Attachment #1 - Page 1 of the CPDETLS-4000 datasheet. Shows the device's features and applications as described by the manufacturer.


Attachment #2 - Illustration of the spreadsheet I used to collect the data intended to characterize the CPDETLS-4000 at 10MHz. It contains the tabulated values obtained during the experiment as well as several calculated values. The notes at the bottom explain how the tabulated values were obtained and show the formulas used for the calculated values.

The reason why I chose to obtain the tabulated values in the manner indicated by the notes is as follows: First, in my opinion, the calibration points given by the CPDETLS-4000 datasheet represent the values that one would see in a true 50 ohm system. To help me figure out the proper signal amplitude required to check a particular calibration point I used an SMA T to connect CH1 of my AWG to the CPDETLS-4000 and to CH1 of the DSO (see attachment #3). Then I varied the AWG's CH1 amplitude until the CPDETLS-4000 DC output matched the DC value given in the datasheet for that particular calibration point. However, the CPDETLS-4000 doesn't have a consistent 50 ohm impedance, so the DSO's CH1 mVpp readings are not suitable for checking the accuracy of the calibration points. To get around this, once the AWG's CH1 was set to the proper amplitude I copied its settings to CH2. I then connected the AWG's CH2 to the DSO's CH2 via a 50 ohm feed-through terminator (see attachment #3). The terminator provides a consistent 50 ohm impedance, so the DSO's CH2 mVpp readings are suitable for checking the accuracy of the calibration points.


Attachment #3 - CPDETLS-4000, DSO, AWG, and DMM connections.

[G0HZU]

Dunno if this helps but I have a HP 8473C detector here that works up to 26GHz. It looks very similar to the one you have and it uses a simple Schottky diode. However, it is a very expensive device because the diode is packaged to work well up to 26GHz. But apart from this I think the technology is similar to the detector you are using.

http://cp.literature.agilent.com/litweb/pdf/5952-8299.pdf

If you look at the spec for input VSWR and frequency response this detector is very flat indeed up to many GHz.

I also have a couple of Agilent ESGD 4433 (4000) sig gens and the typical flatness spec to 4GHz is very impressive. See below for the datasheet spec for a typical generator of this type.

If I connect the 8473C to the sig gen at 100MHz and set the level down at -20dBm this gets me close to square law operation. If I do this the 8473C detector output is about 6.6mVDC. If I turn up the level 1dB it goes to about 8.1mV and if I turn it down 1dB it goes down to 5.25mV DC.

If I then sweep across the full 4GHz range of the sig gen I see very little change in the detector voltage it stays at 6.6mV and probably only changes by +/- 0.1mV. So the detector flatness is very good indeed.

However, if I cool or heat the 8473C then the detector voltage changes quite a bit. i.e. it's only flat and consistent at a fixed temperature when used down at these tiny signal levels.

So it's great for checking flatness but not so great at measuring absolute power levels.

Your detector should be similar although the performance in terms of flatness and input VSWR etc will be dictated by the diode package and also by how well it is fitted inside the body of the detector in terms of stray inductance etc. If this is done with a diode package that has significant stray inductance then you might only get good flatness performance up to 500MHz or maybe 1GHz.

[TomC]

Dunno if this helps but I have a HP 8473C detector here that works up to 26GHz. It looks very similar to the one you have and it uses a simple Schottky diode. However, it is a very expensive device because the diode is packaged to work well up to 26GHz. But apart from this I think the technology is similar to the detector you are using.

http://cp.literature.agilent.com/litweb/pdf/5952-8299.pdf

If you look at the spec for input VSWR and frequency response this detector is very flat indeed up to many GHz.

I also have a couple of Agilent ESGD 4433 (4000) sig gens and the typical flatness spec to 4GHz is very impressive. See below for the datasheet spec for a typical generator of this type.

If I connect the 8473C to the sig gen at 100MHz and set the level down at -20dBm this gets me close to square law operation. If I do this the 8473C detector output is about 6.6mVDC. If I turn up the level 1dB it goes to about 8.1mV and if I turn it down 1dB it goes down to 5.25mV DC.

If I then sweep across the full 4GHz range of the sig gen I see very little change in the detector voltage it stays at 6.6mV and probably only changes by +/- 0.1mV. So the detector flatness is very good indeed.

However, if I cool or heat the 8473C then the detector voltage changes quite a bit. i.e. it's only flat and consistent at a fixed temperature when used down at these tiny signal levels.

So it's great for checking flatness but not so great at measuring absolute power levels.

Your detector should be similar although the performance in terms of flatness and input VSWR etc will be dictated by the diode package and also by how well it is fitted inside the body of the detector in terms of stray inductance etc. If this is done with a diode package that has significant stray inductance then you might only get good flatness performance up to 500MHz or maybe 1GHz.

Thanks for the input and for taking the time to test the 8473C!

One big difference between the 8473C and the CRYSTEK detector is that the Agilent datasheet for their series of LBSD detectors is thorough and answers all pertinent questions, in contrast, the CRYSTEk datasheet leaves way too much room open for speculation.

http://www.crystek.com/microwave/spec-sheets/rfdetector/CPDETLS-4000.pdf

For example, for the 8473C the Agilent datasheet gives an operating temperature of -20°C to +85°C, then at the end of the table there is a note that states: "Above specifications are at 25°C and <= -20 dBm unless otherwise specified.". In contrast, the CRYSTEK datasheet gives an operating temperature of -20°C to 70°C, but it remains mute as to any temperature restrictions for the table of calibration point values given on page 2. Maybe this table, or part of it, is only valid at 25°C or a slightly lower temperature. If that's the case, since the experiment described on my previous post was done at around 28°C, the dismal results at power levels below 0dBm should have been expected!

Although the 8473C and the CPDETLS-4000 both use LBSD detectors, looking at the data sheets it seems to me that there aren't very many additional similarities. As you stated, the 8473C has impressive broadband flatness, frequency response and low VSWR, all of which makes it great for checking flatness. As far as measuring absolute power, I think it can probably also do a good job, but the user must have the appropriate test equipment to calibrate it.

In contrast, the CPDETLS-4000, as evidenced by the graph and table on page 2 of the datasheet, has lousy flatness, and as far as VSWR is concerned, there is no spec. However, the fact that during my experiment it seemed to exhibit > 60 ohms impedance at 10MHz gives some insight as to how lousy its VSWR specs may be. The one redeeming factor, in my opinion, is that the datasheet suggests that the CPDETLS-4000 can be used to measure input power in the range of -10dBm to 10dBm by merely referencing the calibration points table given on page 2. But even this one thing is a riddle, since the datasheet doesn't provide any guidance as to the degree of accuracy that can be expected. Judging from my experiment results, it seems that the window of usable accuracy is limited to input power in the range of 0dBm to 10dBm, at least this seems to be true at 10MHz. So, unless appropriate test equipment is available to calibrate it, it seems to me that readings in the square law region will have too much uncertainty and therefore will not be very useful.

Since I don't own the appropriate test equipment to calibrate this thing, it seems that I'm left with a device that can only provide usable readings while operating in the linear region. As you pointed out earlier, this should work fine if the input signal is a clean sine wave, but where harmonics are present the uncertainty can potentially become large enough to make readings useless. So I want to do a little more research and conduct some experiments regarding how harmonics affect the behavior of LBSD detectors under different circumstances. First I'd like to simulate a harmonic that varies in phase in respect to the fundamental, for example, as the harmonics associated with AM modulation could be expected to behave. Then I'd like to experiment with square wave harmonics, whose phase is fixed with respect to the fundamental. I suspect that in the case of square waves, although the reported dBm level will be incorrect, the corresponding Vpp value will be correct or close to it.

[G0HZU]

So I want to do a little more research and conduct some experiments regarding how harmonics affect the behavior of LBSD detectors under different circumstances.

I've done this a few times in the past. I've also tested popular logamps like the AD8307 for similar issues. The AD8307 is a popular device to use as a power meter up to a few hundred MHz. However, it does suffer significant uncertainty from harmonics if the phase is rotated. It's 15 years or more since I played with the AD8307 at my place of work like this but I recall that it is odd order harmonics that upset it the most.

Diode detectors like my 8473C are affected by both even and odd harmonics when used in the linear region although not as much as theory suggests for a perfectly linear diode. The second harmonic causes more issues with it than the third but both are problematical if you are after low uncertainty.

If it helps, have a read of this classic old app note from HP. It's the one I used many years ago. I think you will find the whole document to be interesting and it may give you a few ideas. But have a look at page 23 onwards and this shows how to model a diode detector wrt the second harmonic as it is used across the square law and linear regions. Up to +/- 20% (or approx +/- 0.8dB) uncertainty in reported power level for a -20dBc harmonic! I don't think your detector will be quite this bad unless you use it way up near its upper limits of range in the linear region.
You may find the uncertainty window becomes very asymmetric (wrt phase angle) at high harmonic levels as well.
So your research results may show a one sided 'pointy/dippy' effect in the uncertainty window rather than a regular +/- ripple effect as you rotate the phase through 360 degrees.

http://www.hparchive.com/seminar_notes/Pratt_Diode_detectors.pdf

Hope it helps...

[TomC]

I've done this a few times in the past. I've also tested popular logamps like the AD8307 for similar issues. The AD8307 is a popular device to use as a power meter up to a few hundred MHz. However, it does suffer significant uncertainty from harmonics if the phase is rotated. It's 15 years or more since I played with the AD8307 at my place of work like this but I recall that it is odd order harmonics that upset it the most.

Diode detectors like my 8473C are affected by both even and odd harmonics when used in the linear region although not as much as theory suggests for a perfectly linear diode. The second harmonic causes more issues with it than the third but both are problematical if you are after low uncertainty.

If it helps, have a read of this classic old app note from HP. It's the one I used many years ago. I think you will find the whole document to be interesting and it may give you a few ideas. But have a look at page 23 onwards and this shows how to model a diode detector wrt the second harmonic as it is used across the square law and linear regions. Up to +/- 20% (or approx +/- 0.8dB) uncertainty in reported power level for a -20dBc harmonic! I don't think your detector will be quite this bad unless you use it way up near its upper limits of range in the linear region.
You may find the uncertainty window becomes very asymmetric (wrt phase angle) at high harmonic levels as well.
So your research results may show a one sided 'pointy/dippy' effect in the uncertainty window rather than a regular +/- ripple effect as you rotate the phase through 360 degrees.

http://www.hparchive.com/seminar_notes/Pratt_Diode_detectors.pdf

Hope it helps...

Thanks for the reply and link!

As it turns out I have already read this article and found it quite interesting, particularly the model you refer to on page 26. In fact, the experiment that I'm working on at the moment is based on this model. I've seen the associated graph on a couple other articles, I believe one was from Agilent, but apparently they had altered the formulas and I couldn't make them work. However, the formulas on this original vintage article by Pratt worked just fine for me! The only thing is the percentage error that he quotes, ±20%, I see that there is about 20% between 0.83dB and -9.92dB if you relate this back to power input levels, but then I come up with about ±10%.

For example, if you are expecting 0dBm (632.5mVpp) but you could end up with as much as 0.83dBm (695.9mVpp) or as little as -0.92dBm (568.9mVpp), then I think you would calculate the percentage error as follows:

   (695.9 - 632.5) / 632.5 x 100 = 10.02%
   (568.9 - 632.5) / 632.5 x 100 = -10.06%

So as I'm working on my experiment I'm still wondering if I'm missing something silly or if there is actually a typo on the article and the correct figure is ±10%.

[G0HZU]

In the linear region the detected voltage will be the true peak of the overall waveform if the diode was perfect and we ignore any Vdrop in the diode for the sake of simplicity.

So if we look at a simple theoretical case for a very large signal being fed to a suitable linear detector diode and the Vpk of the carrier was 10Vpk and the -20dBc harmonic was 1Vpk then the worst case uncertainty one way would be 11V peak (additive) and the other would be 9V peak (subtractive).

power = (Vpk*Vpk)/100 for a 50R system if you assume a perfect sine wave. (but we know this ISN'T a perfect sine wave)

So the uncertainty window would be a falsely indicated power level of  81/100W to 121/100W  or 0.81 to 1.21W as the phase is changed.

This is roughly a (reported) power error of +/- 20%  or -0.92dB to +0.83dB with the -20dBc harmonic present.

Obviously, the detector you are using can't cope with voltages this big but I've used big numbers to demonstrate linear behaviour so the numbers are easy to deal with.

[TomC]

In the linear region the detected voltage will be the true peak of the overall waveform if the diode was perfect and we ignore any Vdrop in the diode for the sake of simplicity.

So if we look at a simple theoretical case for a very large signal being fed to a suitable linear detector diode and the Vpk of the carrier was 10Vpk and the -20dBc harmonic was 1Vpk then the worst case uncertainty one way would be 11V peak (additive) and the other would be 9V peak (subtractive).

power = (Vpk*Vpk)/100 for a 50R system if you assume a perfect sine wave. (but we know this ISN'T a perfect sine wave)

So the uncertainty window would be a falsely indicated power level of  81/100W to 121/100W  or 0.81 to 1.21W as the phase is changed.

This is roughly a (reported) power error of +/- 20%  or -0.92dB to +0.83dB with the -20dBc harmonic present.

Obviously, the detector you are using can't cope with voltages this big but I've used big numbers to demonstrate linear behaviour so the numbers are easy to deal with.

Thanks, that cleared it up!

What I was overlooking in my calculation is that when you convert from dBm to Vpp you also have to consider the impedance when talking about power levels. So I correctly found the percentage error for the voltage levels, not the power levels, and although I went over it several times I still wasn't seeing it. Once I saw your example it hit me like a flash of lighting, well, at least I figured that I was probably missing something silly!

So I think that my example would be correct If I add the impedance: If you are expecting 0dBm (632.5mVpp @ 50 ohms = 1mW) but you could end up with as much as 0.83dBm (695.9mVpp @ 50 ohms = 1.21mW) or as little as -0.92dBm (568.9mVpp @ 50 ohms = 0.81mW), then you can calculate the voltage percentage error as follows:

   (695.9 - 632.5) / 632.5 x 100 = 10.02%
   (568.9 - 632.5) / 632.5 x 100 = -10.06%

and the power percentage error as follows:

  (1.21 - 1) / 1 x 100 = 21%
  (0.81 - 1) / 1 x 100 = -19%

So Thanks again G0HZU, that really helped!

[TomC]

This post describes the first of a series of experiments where I try to explore the effect of harmonics on the behavior of LBSD detectors such as the one used in the CPDETLS-4000. On this first experiment I'd like to look at harmonics that vary in phase in respect to the fundamental, for example, as the harmonics associated with AM modulation could be expected to behave.

The experiment I have in mind came to me after reading "Characteristics and Applications of Diode Detectors" by Ron Pratt:

www.hparchive.com/seminar_notes/Pratt_Diode_detectors.pdf

On page 26 of this document there is a graph titled "Error Produced by 2nd Harmonic Predicted by Simple Model", see attachment #1. The experimental test described below is meant to help visualize with actual waveforms and absolute values the analysis described on this page.

The steps below describe how I simulated a time domain signal with the characteristics of the frequency domain graph illustrated on the left hand side of #1. This is then used to help predict the behavior of a perfect LBSD detector operating in the linear region (No junction voltage drop and an output voltage that corresponds perfectly to the peak voltage of the input signal) when this signal is applied to its input:

   1. I used my dual channel AWG to generate two signals, a 10MHz 2Vpp (10dBm) sine wave used to represent the fundamental and a 20MHz 200mVpp (-10dBm) sine wave used to represent a 2nd harmonic. The -20dBc ratio was chosen so it would coincide with the model depicted by attachment #1.

   2. I fed the AWG's output signals to CH1 & CH2 of my DSO and used the DSO's math function to add the fundamental and the second harmonic. The resulting trace (green trace on my DSO) represents the way that the combined signals would appear at the detector's input.

   3. While observing the DSO's green trace, I slowly changed the phase of the harmonic in respect to the fundamental with the AWG's phase control. The high and low extremes that corresponds to the graph on the right hand side of #1 occur when the positive peaks are in phase, and when the negative peaks are in phase, see attachment #2.

Assuming that the power detector is operating in the linear region, uses a single diode arrangement, and the input signal is connected directly to the diode, then the detector's output should be very close to the peak voltage of the green trace. The peak voltage of a signal can be used to accurately derive its power level if the signal is a clean sine wave, for example, if the harmonic wasn't present the peak voltage in this case would be around 1V and the dBm level could be calculated with the formula:

   dBm = 10log_10[(Vpp)^2 / (0.008R)] = 10log_10[(2^2 / 0.4)] = 10dBm.

However, with the harmonic present the extremes produce 1.10V and 0.9V respectively (see #2). So the reported dBm levels are:

   dBm = 10log_10[(2.2^2 / 0.4) = 10.83dBm
   dBm = 10log_10[(1.8^2 / 0.4) = 9.085dBm

Since the expected power level is 10dBm (10mW), but we could end up with as much as 10.83dBm (12.1mW), or as little as 9.085dBm (8.1mW), the power percentage error can be determined as follows:

   (12.1 - 10) / 10 x 100 = 21%
   ( 8.1 - 10) / 10 x 100 = -19%

And the voltage percentage error can be determined as follows:

   (1.1 - 1) / 1 x 100 = 10%
   (0.9 - 1) / 1 x 100 = -10%

The above values represent the worst case error caused by the presence of the harmonic. The uncertainty range in dB can be determined by subtracting the expected value from each extreme:

   10.83dBm - 10dBm = 0.83dB
   9.085dBm - 10dBm = -0.92dB

Alternatively, it could be determined by converting the ratio of the power levels to dB:

   dB = 10log_10(P_1 / P_2)
   dB = 10log_10(12.1 / 10) = 0.83dB
   dB = 10log_10( 8.1 / 10) = -0.92dB

These simulation results perfectly coincide with the values given in #1.


Attachment #1 - Graph of the error produced by a 2nd harmonic depending on the region of operation of the diode detector. This came from an HP document written by Ron Pratt. I've seen almost identical graphs on other documents but I couldn't verify the cited values with the given formulas. However, the formulas given in this version worked out just fine for me.


Attachment #2 - Simulation of the effect of a 2nd harmonic on the fundamental starting when the positive peaks are in phase and ending when the negative peaks are in phase. Channel 1 of the AWG is set to 10MHz 2Vpp, Channel 2 is set to 20MHz 200mVpp. Each AWG channel is connected to the corresponding DSO channel via a feed-through 50 ohm terminator. The green trace was produced by adding the fundamental (red trace) to the harmonic (yellow trace) using the DSO's math function. The AWG's phase control allows the user to adjust the phase from 0 to 360 degrees.

-------------------------------------------------------------------------------------------------------

Next I'll try to combine these signals using a BNC T and connect them to the CPDETLS-4000.


Edit: I wasn't happy with the way Attachment #2 looked when displayed on my PC monitor. The resolution wasn't the same as when I tried it off-line, so I added single image attachments #2a & #2b for anyone having difficulty viewing the details on attachment #2.

[TomC]

This post is the sequel to post #20. Its the second of a series of experiments where I try to explore the effect of harmonics on the behavior of LBSD detectors such as the one used in the CPDETLS-4000. On this experiment I use a standard BNC T to combine the signals from the two AWG channels. See post #20 for details of how the AWG was setup.


Attachments #1 & #2 - Here the BNC T is directly connected to the DSO's CH1. The AWG is setup the same as it was for post #20. Since the AWG channels terminate each other via the BNC T, there is no need for the 50 ohm feed-through terminators used on post #20. The CH1 display is nearly identical to the green trace produced by the DSO's math function on post #20. For calculations I'm using the values from the measurements at the bottom left of the screen instead of the cursors used on post #20. On this DSO Vp = Vpp, Vt = Vtop, and Vk = Vrms.

As in the previous post, if the harmonic wasn't present the peak voltage in this case would be around 1V and the dBm level could be calculated with the formula:

   dBm = 10log_10[(Vpp)^2 / (0.008R)] = 10log_10[(2^2 / 0.4)] = 10dBm.

However, with the harmonic present the extremes produce 1.10V and 0.9V respectively as indicated by Vt. So the reported dBm levels are:

   #1. dBm = 10log_10[(2.2^2 / 0.4) = 10.83dBm
   #2. dBm = 10log_10[(1.8^2 / 0.4) = 9.085dBm

Since the expected power level is 10dBm (10mW), but we could end up with as much as 10.83dBm (12.1mW), or as little as 9.085dBm (8.1mW), the power percentage error can be determined as follows:

   #1. (12.1 - 10) / 10 x 100 = 21%
   #2. ( 8.1 - 10) / 10 x 100 = -19%

And the voltage percentage error can be determined as follows:

   #1. (1.1 - 1) / 1 x 100 = 10%
   #2. (0.9 - 1) / 1 x 100 = -10%

The above values represent the worst case error caused by the presence of the harmonic. The uncertainty range in dB can be determined by subtracting the expected value from each extreme:

   #1. 10.83dBm - 10dBm = 0.83dB
   #2. 9.085dBm - 10dBm = -0.92dB

Since the DSO provides a Vrms value for CH1 we can also calculate the "actual" dBm value of the combined signal:

   dBm = 10log_10[(Vrms)^2 / (0.001R)]
   #1. dBm = 10log_10[0.7178^2 / 0.05] = 10.13dBm
   #2. dBm = 10log_10[0.7165^2 / 0.05] = 10.11dBm

Note: In theory the Vrms value should be the same regardless of the phase but since the DSO reports slight differences I used the exact reported values for the calculations.


Attachments #3 & #4 - Here the BNC T is connected to the DSO's CH1 via a 50 ohm feed-through terminator. The AWG is setup the same as for attachments #1 & #2 except that the "load" setting is set to 25 ohms instead of 50 ohms. This is necessary because each channel was already seeing a 50 ohm load via the BNC T, the 50 ohm feed-through is in parallel so the load seen by each channel is now 25 ohms. Note that the DSO's display is nearly identical to what we got for #1 and #2 except for a slight difference in the Vrms values, again, in theory, these should all be the same. However, for completeness sake, here are the calculations:

   #3. dBm = 10log_10[0.7197^2 / 0.05] = 10.15dBm
   #4. dBm = 10log_10[0.7188^2 / 0.05] = 10.14dBm

With this step of the experiment the only thing we achieved is to verify that the AWG setup behaves as expected in a 50 ohm system. On the next step we'll remove the 50 ohm feed-through terminator and connect the CPDETLS-4000 in its place. If the DSO reports any level changes we'll know that is not the AWG settings and that the reason is that the CPDETLS-4000's impedance is not quite 50 ohms.


Attachments #5 & #6 - Here the BNC T is connected to both CH1 of the DSO and to the CPDETLS-4000 via an SMA T. The CPDETLS-4000 DC output is connected to the DMM via a short RG316 cable. For this step of the experiment the DSO connection is not essential, we are mostly interested on the CPDETLS-4000's DC output. However, its interesting to see if the DSO reports different signal levels with this connection just to assess how well the detector's input matches the system's impedance. In addition, it helps verify that the AWG phase setting is correct for the readings at the two extremes.

   #5. CPDETLS-4000's DC output = 787.8mV reported by the DMM.
       This translates to 2150.5mVpp or 10.63dBm as per the datasheet's calibration points.

       Since the DSO reports signal levels > #3, this means that the CPDETLS-4000's impedance is > 50 ohm.
       These signals are not used for any other calculations
 
   #6. CPDETLS-4000's DC output = 614.1mV reported by the DMM.
       This translates to 1764.1mVpp or 8.91dBm as per the datasheet's calibration points.

       Since the DSO reports signal levels > #4, this means that the CPDETLS-4000's impedance is > 50 ohm.
       These signals are not used for any other calculations
 
Now let's go through the calculations we did for #1 & #2 again so that we can compare the behavior of the CPDETLS-4000 to the perfect detector previously envisioned.

Again, if the harmonic wasn't present the expected peak voltage would be around 1V and the expected dBm level is calculated with the formula:

   dBm = 10log_10[(Vpp)^2 / (0.008R)] = 10log_10[(2^2 / 0.4)] = 10dBm.

Note that 10dBm is the level that we expect from the perfect detector. This is probably different from what you would get with the CPDETLS-4000. From previous experiments we know that it under reports at this power level, so it would probably report around 9.9dBm. However, for the comparison calculations below we want to use the same level used for the perfect detector.

With the harmonic present the extremes produce 2.1505Vpp and 1.7641Vpp respectively as indicated by the datasheet's calibration points. So the reported dBm levels are:

   #5. dBm = 10log_10[(2.1505^2 / 0.4) = 10.63dBm (compared to: 10.83dBm)
   #6. dBm = 10log_10[(1.7641^2 / 0.4) =  8.91dBm (compared to:  9.085dBm)

Since the expected power level is 10dBm (10mW), but we could end up with as much as 10.63dBm (11.6mW), or as little as 8.91dBm (7.8mW), the power percentage error can be determined as follows:

   #5. (11.6 - 10) / 10 x 100 =  16% (compared to:  21%)
   #6. ( 7.8 - 10) / 10 x 100 = -22% (compared to: -19%)

And the voltage percentage error can be determined as follows:

   #5. (1.075 - 1) / 1 x 100 =   7.5% (compared to:  10%)
   #6. (0.882 - 1) / 1 x 100 = -11.8% (compared to: -10%)

The above values represent the worst case error caused by the presence of the harmonic. The uncertainty range in dB can be determined by subtracting the expected value from each extreme:

   #5. 10.63dBm - 10dBm = 0.63dB (compared to:  0.83dB)
   #6. 8.91dBm - 10dBm = -1.09dB (compared to: -0.92dB)

Since the DSO provides a Vrms value for CH1 we can also calculate the "actual" dBm value of the combined signal:

   #3. dBm = 10log_10[0.7197^2 / 0.05] = 10.15dBm
   #4. dBm = 10log_10[0.7188^2 / 0.05] = 10.14dBm

As indicated, these "actual" dBm values came from the DSO readings obtained on #3 & #4. As previously stated, We can't use the DSO readings from #5 & #6 for any of these calculations, they are inaccurate due to the CPDETLS-4000's impedance.


Some things about the CPDETLS-4000 that I learned or verified with this experiment:


   1. It uses a single diode arrangement. This is evident because there is a large difference in reported values between the extreme where the positive peaks are in phase and the extreme where the negative peaks are in phase.

Some power detectors use two diodes in a push-pull arrangement. In this case the reported value is the average of the positive and negative peaks, so there wouldn't be a significant difference between the extremes. One advantage of this dual diode arrangement is that even in the linear region, when even harmonics are present the reported dBm value is much closer to the "actual" value. For example, in the case of #3 & #4 the reported value would have been about 1/2 the Vpp or around 1.01V. So the dBm value would be:

   dBm = 10log_10[(Vpp)^2 / (0.008R)] = 10log_10[(2.02^2 / 0.4)] = 10.09dBm

Much closer to the "actual" dBm value which my DSO reports as somewhere between 10.11dBm and 10.15dBm.


   2. The diode is wired so that it detects the positive peaks. This is evident because the output DC level is always positive. In addition, this level is higher for the extreme where the positive peaks are in phase.

Some power detectors are wired so that they detect the negative peaks, in some cases this a feature that can be specified when the device is ordered.


   3. The input signal is not directly connected to the diode. This is evident because there is a large difference between the output DC level and peak voltage of the input signal. To me this indicates that there is significant attenuation before the signal is applied to the diode, possibly caused by the components of an impedance matching network.


---------------------------------------------------------------------------------------------------------


Next I plan to do some experiments involving square waves.

[TomC]

When I started this topic one of the main objectives was to measure and report the output levels of the SynthUSBII at different frequencies. At the time I didn't have appropriate equipment to get accurate results. Now that I have a Spectrum Analyzer, I decided to add that last bit of information to this topic. The SynthUSBII was directly connected to the SA's input (no cables) when the readings in the attached photos were obtained. The four traces represent the 4 output levels available on the SynthUSBII.