Digital Attenuators - your experiences , suggestions and comments

[noreply]

Ok, I finally managed to do some frequency response testing on the digital attenuator module.

To do a very thorough test – is not an easy task and would take a very long time to acquire and more importantly present the data in a meaningful way.

So, I decided to do some ‘hands-on’ testing to see how effective the device is for real day to day use and how well it behaves in relation to its ‘claimed’ specifications.

Since the actual attenuation ‘chip’ has been defaced (etched) so that it cannot be identified – possibly because it’s an inferior specification ‘clone’ , I am assuming that the device which ‘should’ be inside is this;-

https://www.psemi.com/pdf/datasheets/pe43713ds.pdf


Spoiler alert


The results obtained – don’t reflect the specifications in the above data sheet.


However – there is a relatively linear response up to about 700MHz

AND
 
There is also ‘expected’ (close to linear) behavior at the 3.1GHz point for most of the attenuation settings.


This, together with a closer look at the Evaluation Kit Layout for PE43713 (see attached pictures) – will reveal that the PCB used in the evaluation design is nothing like the POOR quality of the actual module under test.

You will note that the RF line microstrip  for the input and output on the Evaluation Kit Layout for PE43713 is precise with a series of through vias to the ground plane – which will ensure a better response at higher frequencies.

I suspect that even if the semiconductor used is NOT a clone – its PCB layout is responsible for the poor frequency response.

Further experimentation could ultimately verify this.

As a possible confirmation to the above theory – I made a frequency response plot with NO POWER to the device.

To some expend this should give a ‘loose’ indication of the ‘transmission path’ response within the device under test.

You can see that the frequency response plot follows the same ‘peaks’ and ‘thoughts’ as does the device when operations – a small indication that there is a definite problem with the signal path before the device is even ‘powered-on’.


Anyway, I made a few assumptions for the testing.


The tracking generator was set to 0dBm output

The sweep frequency was from 0Mhz to 3.2Ghz

I set the marker to catch the highest peak – so we can see the ‘lowest’ attenuation figure when set to a predetermined value – as this will be more critical than having a higher attenuation figure if this attenuation module is used in some design under test.

The interesting thing is that the ‘lowest’ – most true to the setting – attenuation occurs at 3.02GHz  - sorry I was unable to see what happens beyond the 3.2 GHz as my SVA does not cover above 3.2GHz – perhaps someone else with same device could perform higher frequency testing?

At -20dB attenuation – the module had a ‘peak’ response at 776.53MHz – where it reported an attenuation value of -21.21 dB – this figure is within the PSEMI PE43713 device specifications – again a possible clue that the PCB layout used in this module is killing the performance and not the ‘chip’ itself?

You can see the attenuation vs. frequency is relatively linear up to 700MHz – yet another ‘clue’ to poor PCB design as frequencies above this and beyond 1GHZ definitely need to take into account the PCB dielectric and the PRECISE microstrip design calculations in relation to the copper tracks used in the layout.


I also did not make any deviations on the TG power level – all measurements at 0dBm


I suspect that using variable input power would have ‘shifted’ some of the frequency and attenuation values – as the behavior of the ‘transmission’ of the RF signal would change with the higher power.


Perhaps I will be able to do this test later – if some of you think is would help?


My guess, we have enough data at the 0dBm level to get a good feel as to how this device behaves.


I also noted that the displayed frequency on the OLED display is ‘fixed’ at the 6.0GHz Level – irrespective of what the input frequency into the attenuator is set to!

I feel this is misleading advertising – I had the impression that the device would show the Frequency of the attenuated signal – a useful feature to have – but obviously not supported by the PSEMI chip – so no way to implement.

Then why have it??

Could display some other meaningful data or leave the 2nd line on the OLED blank.



Conclusion


Would I buy this device?

For myself – probably not – because for me an attenuator’s value is CRITICAL in all of my circuit designs and devices under test.

I need to ‘trust’ that the attenuator will provide a ‘flat’ response within its specifications and at a constant attenuation level.

My alternative – although not as convenient as a digital module – would be to use a SMA Attenuator Kit - Bundle of 6pc 2W 50 Ohm SMA In-Line Attenuators which Provides Highly Linear Attenuation from 1dB to 42dB in 1dB Increments.  With the following individual adaptors 1dB, 2dB, 3dB, 6dB, 10dB & 20dB

This set of attenuators will cost about GBP41 compared to GBP32 for the module under review.



Is this device useful?

Yes – potentially to someone who will use this at lower frequencies – say between 10MHz and 100MHz – where it will provide linear response – close to published specifications – but NOT precise to the level of fixed attenuators.



Can this module be ‘fixed’?


Yes, I suspect this would be a nice project to ‘fix’

The microcontroller module is on a separate PCB attached with a through header set of pins – linking the two PCB’s together.

You would need to remove the main PCB with the PSEMI PE43713 device (I am assuming it’s not a clone chip) and redesign the PCB – or use the layout (section of) the Evaluation Kit Layout – with appropriate PCB substrate.

If my theory is correct – you will then be able to have a much better performing digital attenuator – working within the published datasheet specifications – which are quite impressive.




Hope that above ‘quick and dirty’ review is useful in your decision in purchasing this module 

Be warned – out of 3 units purchased – 2 had OLED display failures – one was DOA and the other failed within 3 minutes of being powered-on.

The packaging used for these modules is VERY BAD – it is highly likely that some damage will occur if the anti static package containing the attenuator module and cable is not wrapped in lots of bubble wrap and not simply put inside an Amazon type large cardboard envelope.

If you decide to purchase, hope you have better luck than me –– and when your device arrives it is not DOA or damaged during transit.


If you are still reading this last line - thanks for your patience. 

[edigi]

I also noted that the displayed frequency on the OLED display is ‘fixed’ at the 6.0GHz Level – irrespective of what the input frequency into the attenuator is set to!

I feel this is misleading advertising – I had the impression that the device would show the Frequency of the attenuated signal – a useful feature to have – but obviously not supported by the PSEMI chip – so no way to implement.

Then why have it??

I assume that at some point of time they had such ambitious plan that the attenuation is compensated with measurement results possibly stored in the micro (maybe not in this, but a more expensive device) 
For that the micro has to know which frequency to compensate. 
This could be given via USB command or as a very unlikely case measured by the device itself.
Probably I'm assuming way too much here as it shows only the upper frequency limit currently which is blatantly misleading (and will create burn in the OLED module pretty soon - from my personal experience this kind of modules are very prone to it).

My alternative – although not as convenient as a digital module – would be to use a SMA Attenuator Kit - Bundle of 6pc 2W 50 Ohm SMA In-Line Attenuators which Provides Highly Linear Attenuation from 1dB to 42dB in 1dB Increments.  With the following individual adaptors 1dB, 2dB, 3dB, 6dB, 10dB & 20dB

I have more or less that kind of SMA attenuator series (although shifted right so having also 40 and 50dB), in some situations they are better than the electronic one, sometimes the electronic one could be simpler to use.

The microcontroller module is on a separate PCB attached with a through header set of pins – linking the two PCB’s together.

You would need to remove the main PCB with the PSEMI PE43713 device (I am assuming it’s not a clone chip) and redesign the PCB – or use the layout (section of) the Evaluation Kit Layout – with appropriate PCB substrate.

If my theory is correct – you will then be able to have a much better performing digital attenuator – working within the published datasheet specifications – which are quite impressive.

The reason why I've bought just a PCB without any extra stuff (basically just a breakout board of the chip) was pretty much along these lines (if you check my earlier photo it even seems to have proper via stitching).
Just using the datasheet I was able to control dozens of various devices since they have typically some kind of SPI interface and reasonable control.
I've burned myself once with these OLED stuff, never again. (That was a frequency counter 1 out of 2 working but with horrendous resolution and accuracy - if we can talk about any of these, since it was 5 magnitudes of order below of hobbyist grade stuff and I'm not kidding here). 
So far I have not done this (still not entirely abandoned the idea though) but if those micro switches on my PCB will break maybe it will create a new surge to this project.

[edigi]

I've decided to make a quick test of my PE43702 breakout board (this is a 4GHz device, not 6GHz) at max attenuation.

For fun, I've done the testing with both Siglent SSA and NanoVNAv2. Both set to 5dB/div calibrated with everything the same (as far as it was possible).
Due to that this is just unshielded PCB, I can influence the result especially in the higher frequency range with a mere touch of the ground of the SMA connector...
The big deviation at low frequencies is due to the capacitors at the input/output (I also suspect that they don't improve the flatness of the curve...).

In the NanoVNAv2 thread I've posted some test results with SMA attenuators, should you want to compare with them.
https://www.eevblog.com/forum/testgear/nanovna-v2-aka-s-a-a-2/msg3137696/#msg3137696

[noreply]

Interesting ...

I guess for a low cost device - the NanoVNA is OK to give you the 'trend' - but hard to get actual measurements from this device compared to a proper SSA.

Your NanoVNA - definitely shows similar response as SSA.

Your SSA response if VERY similar to the device I tested - maybe the 'semiconductor' - since they etched the markings - is same for ALL digital attenuator modules coming out of China via eBay and Amazon because its a cheaper clone version rather than original PSEMI device?

In your SSA frequency response - you get a 'peak' at about 3 GHz - same as I did, its also relatively 'flat' up to 700MHz.

I'm not sure if your PCB layout is same as the module I tested (there is a photo of the PCB in an earlier post) - if it is - then this could be the reason, it it is NOT same - then my PCB layout theory might not be so 'strong' in trying to improve the quality of the resultant attenuated signal.

If your board has some additional vias - or looks similar to the Evaluation Kit Layout for PE43713 (photo of layout posted in thread above) - then this will be a big clue that the 'response' is most likely 'chip' related and not PCB layout or microstrip - lack off - related.

With your results - on a different design to what I tested - my guess is that its 'chip' related - probably why they scrubbed the marking off the semiconductor 

There is still some millage left in this investigation - if you are willing to 'enhance' your PCB layout to see if it can temper the resultant frequency response to that closer to the real device's published specifications?

I would have done some experimentation with my device - but I decided to return this instead (after all it had a broken switch) - because its simply not possible to use it for my needs - as I need ACCURATE attenuation values always - throughout the specified frequency range!

Thanks for your valued feedback 

[edigi]

I guess for a low cost device - the NanoVNA is OK to give you the 'trend' - but hard to get actual measurements from this device compared to a proper SSA.

As long as you stay within the limits of NanoVNA (and the PE43702 module that I've tested is comfortably within) it's actually fairly easy to use it for actual measurements. Although I could challenge it with my SMA attenuators, it's still a great value in my view.

I'm not sure if your PCB layout is same as the module I tested (there is a photo of the PCB in an earlier post) - if it is - then this could be the reason, it it is NOT same - then my PCB layout theory might not be so 'strong' in trying to improve the quality of the resultant attenuated signal.

If your board has some additional vias - or looks similar to the Evaluation Kit Layout for PE43713 (photo of layout posted in thread above) - then this will be a big clue that the 'response' is most likely 'chip' related and not PCB layout or microstrip - lack off - related.

With your results - on a different design to what I tested - my guess is that its 'chip' related - probably why they scrubbed the marking off the semiconductor 

There is still some millage left in this investigation - if you are willing to 'enhance' your PCB layout to see if it can temper the resultant frequency response to that closer to the real device's published specifications?

Due to that the forum engine messed up the thumbnail to photo matching I cannot really check any of your photos, but based on even the thumbnail PCBs look different.
Out of curiosity (and possibly to have a better attenuator although even what I have is OK for me till around 1.2-1.4GHz) I've ordered a PE43703 board. Based on the advertisement photos it's more similar to what I have currently, but still different. Let's see if it's any better (probably it takes quite some time till it arrives to me).
Sorry I'm not willing to waste any time on trying to mod the PCB (it's harder than it looks and it has already a usable range).

I include a high res photo of the chip if anyone is able to identify anything based on that (to me it looks like it follows datasheet marking but that's it).

[noreply]

I include a high res photo of the chip if anyone is able to identify anything based on that (to me it looks like it follows datasheet marking but that's it).

At least your PSEMI device looks genuine - it certainly did not have its markings removed - that's a good thing.

Here is a link to an application note for your specific device - to help you 'tune' and play with it 

https://www.psemi.com/pdf/app_notes/an26.pdf

I guess that we are getting 'similar' SVA peeks - would indicate to me that my device in the 6HZz version is probably a genuine PSEMI but most likely same as yours - a 4GHz and not 6GHz device - hence why the scrubbed the markings.

Not every 'dude' who buys this module on eBay will have a $5000 SVA to check if they are getting ripped of with an inferior specification module 

I am attaching my close-up and wide shot of the CHIP and PCB again

*** I just noticed - see p8.jpg ***

I think they tried to do 'on the fly' microstrip design - possibly applying solder to the copper strip as maybe while looking at the SA frequency response??

It certainly look unusual with all the solder 'sculpture'?

What do you think?

I guess its an easy way to experiment 

 

Looking forward to you 'new' feedback - when you receive your new module.

[noreply]

Interesting behavior - UPDATE

My 3rd device purchase via Amazon - was '100%' working - and within 10 minutes the OLED screen went blank 

I presume the 'device' was still functioning - but it was simply not possible to control it via the pushbuttons - when the screen can't be used for letting you know what's happening.


So I put this 'now dead' device aside - scheduled for yet another Amazon return (this was the 3rd  device which was faulty - the one I did the tests on had faulty switch 2nd device - which I managed to bypass for the testing)

Today (some 5 days since powering it on and its ultimate OLED screen failure) I decided to 'power' up the device - just to see if it was really dead or something like a miracle transpired and it working again.

So I flick the switch and 'bingo' the device is fully functional - with the OLED screen working as though nothing happened in the first place.

I decided to stress test it - by making it go through at least a dozen power-on off cycles - all worked - no problems.

So I decided to leave it on - a long term 'burn in' - to see if it lasts longer than 10 minutes as it did last time.

One hour later - its still going strong , but 90 minutes later or thereabouts (i did not have clock) - the screen fails - dead again as before.

OK, so what just happened?

I decided to do a quick 'freeze' dropped the temp of the device by a good 15 dec C - then decided to power-up

Still dead - so not an obvious temperature problem

I  leave the device in 'off state' for about 3 hours - then a quick check - still dead 

So, I'm thinking - this is really unusual?

I am going to leave it for several days - like before and then power-up again to see what happens.

I thought that - this dead then alive - then dead behaviors was a noteworthy observation.

I don't have much experience with OLED displays - just wondering if this is something common to the type of small OLED displays as used in this device?

Where they can come to life after a while and then die again?

Anyway - thought I should share this experiance with you all 

[bicycleguy]

I've had to much wine and must comment.  Why does everyone posting NanoVNA plots leave all kinds of extraneous traces on when you can turn them off?  Who needs a smith chart overlay from something on a LogMag thru plot?  At minimum please say which trace you are looking at.  Nice pictures and observations though.

Sorry for the rant but its been building up over time on almost every NanoVNA plot.  Not great for nubs.

I've decided to make a quick test of my PE43702 breakout board ...

[edigi]

The Smith chart also provides very valuable information here.
You can see from it how well the input impedance of the attenuator is matched.
If you compare it with the SMA attenuator charts, you can see that this attenuator is not very well matched at any frequency. In case of the SMA attenuator it's just a single point (at system impedance = 50 Ohm) thus it's very well matched in the entire range, unlike with this attenuator where it's a curve, mostly capacitive (except at the highest frequency range).

So please forgive me providing complete information for those who are interested and anyhow it's also allowed to be lazy in EEVblog, isn't it. 

Instead of starting a (fake) rant, you could have checked perhaps one of the Smith chart videos from w2aew (he has many excellent tutorials related to Smith charts, but not only that also NanoVNA or IQ modulation and so on).
As a starting point:

[bicycleguy]

Thanks, great video link and explanation.

[noreply]

Today I received yet another module the 4th - in my quest to get a FULLY functional unit.

The good news it that it appears to work - at least for now.

The OLED display is functioning.

I connected the module to my SVA and did a sweep from 0MHz to 3.2GHz (my limit)

See the attached plot.

Sorry - I simply forgot to label each of the traces - its obvious though - I started at 0dB then 10dB, 20dB and finally 30dB

Since these failed so dismally - I did not bother to check the actual accuracy of the attenuation steps - which are adjustable in .05dB steps.

Weather the semiconductor 'chip' will recognize these 0.05dB steps - is debatable - I don't believe from my previous checking. The built-in microcontroller gives you the ability to select such small steps - another lure - to make you think this is a great module to buy and use - but in fact it is deceiving the buyer once again 


The results as almost same as the previous module (with faulty switch). 

This makes me believe that the 'limiting factors' have nothing to do with the individual PCB as the variation in a 'bad' PCB would not produce almost identical response over the entire 3.2GHz span.

So, given that the response is terrible beyond about 700MHz - I would seriously think that the semiconductor used in this module is definitely not specified to operate at the 6GHz upper limit.

The additional information - which supports the 'fake semiconductor' theory - is that ALL modules had the chip markings removed - something they would not do if the correctly specified semiconductor would be used.

I guess - they are relying on the most plausible scenario - where the person buying a $30 module either from China directly , eBay or Amazon is simply not going to know that the module does not work to specifications beyond 700MHz - because they don't have a $5K SVA at their disposal.

The lesson here - buyer beware - don't TRUST these modules to work to specifications - without testing it yourself (if you can).

I am returning this 'working module'  to Amazon - with a copy of the SVA frequency response vs attenuation - asking Amazon to forward this to the seller and make sure they don't make false claims - just to prove my point that not every sale will simply be accepting of false specification claims by the seller.

Hope that my investigations and the above exercise was useful but I guess everyone will have their own experiences 

[edigi]

I guess - they are relying on the most plausible scenario - where the person buying a $30 module either from China directly , eBay or Amazon is simply not going to know that the module does not work to specifications beyond 700MHz - because they don't have a $5K SVA at their disposal.

The lesson here - buyer beware - don't TRUST these modules to work to specifications - without testing it yourself (if you can).

You keep mentioning the $5K SVA needed thing for checking this attenuator, while in reality the 60$ (or so) NanoVNAv2 is more than sufficient to see that you can't expect very high precision from it.
I heavily suspect that even the designer did not have a $5K SVA for checking this thing (not to talk about the seller, the wholesale company and so on; they probably didn't even know what to look for even if there was a decent spec.).
You have to have some sense of reality, RF/microwave components are typically expensive and if something is cheap and if it does not have decent documentation/spec (or some forum where it's extensively discussed) you have to live with the suspicion that it's a hobbyist grade stuff at best.
Nevertheless this does not mean that what is cheap is necessarily bad or what is expensive is necessarily a good value.
This is why I'm usually willing to take risk with not too expensive things (sometimes I look ways to enhance them but not in this category; example is the AWG as many people did that), although I'd have never bought a 3rd and not to talk about a 4th copy of the same thing. Even if the OLED or the power switch or whatever works, the key components like the signal path on the PCB and the attenuator chip is very unlikely to be different...

[tautech]

You keep mentioning the $5K SVA needed thing for checking this attenuator,............

Seems that way don't it but it's not as many are converting their $1.6k SSA3021X+ into a 3.2 GHz SVA. 

[edigi]

Seems that way don't it but it's not as many are converting their $1.6k SSA3021X+ into a 3.2 GHz SVA. 

I think you've entirely missed my point which was that one does not need a $5K SVA (or a $1.6k SVA converted SSA for that matter) when a 60$ NanoVNAv2 is perfectly capable checking the attenuator of this topic (and a lot more).

[noreply]

I guess - they are relying on the most plausible scenario - where the person buying a $30 module either from China directly , eBay or Amazon is simply not going to know that the module does not work to specifications beyond 700MHz - because they don't have a $5K SVA at their disposal.

The lesson here - buyer beware - don't TRUST these modules to work to specifications - without testing it yourself (if you can).

You keep mentioning the $5K SVA needed thing for checking this attenuator, while in reality the 60$ (or so) NanoVNAv2 is more than sufficient to see that you can't expect very high precision from it.

@edigi

You are spot-on with your comment

I don't know what came over me - to simply quote that not everyone has a $5K SVA - so they are unable to 'check' the operation of a cheap attenuator module.

I was one of the early adopters of the nanoVNA - in fact I have 3 of them - they are extremely useful devices - especially for checking antennas and attenuators - just to name two types of measurements / testing.

Just before I performed the Digital attenuator module testing on my 4th device (more on this later) - I was doing some SVA testing of RF signal generators - measuring noise floor levels and harmonic responses - which were beyond the nanoVNA's capability - so I guess I had this 'on my brain' when making the rash remark that you need to have a $5K Test instrument 

You made a good point and I hope those reading this thread - if they get to this post - they should take note

Why did I preserver with getting a 4th module?

I guess I wanted to get a fully functional unit (with no faults , like broken switch and nonworking OLED display) - because I wanted to 'keep' it for use in my lab.

What I did not realize, until much later, that the module performance has little to do with the PCB and or its layout (but there could still be a possibility if the layout is total garbage and in essence creates a reflective transmission line after 700Mhz?) and more likely the semiconductor with its scrubbed out markings to conceal that it is not a 6GHz device.

I probably will keep the device for the following reasons;-

1. It can be used for low frequency applications - up to 500MHz
2. The aluminum RF casing with the SMA connectors is alone worth the $$ - as I often need a 'package' to house my microstrip designs.


Hope the above makes sense to anyone wanting to purchase & test a similar device to the one tested here.

[edigi]

I've received the PE43703 board (see attached picture).
Surprisingly the power plug is reversed, so what is GND in the PE43702 board it's 5V here and vice versa. You must pay special attention to this (although probably there is reverse polarity protection).

I did the testing with NanoVNA.
After seeing the results I've cross checked the NanoVNA with SMA attenuators. With those it's almost perfectly flat.
The good thing with these electronic attenuators that the attenuation can be set in fine steps but the price seems to be that there is frequency dependence (at least with these variants).
What I particularly dislike is the poor impedance matching and the resulting reflections.

[noreply]

@edigi

First of all I prefer your board - to the OLED with microcontroller - using dip switches is far more reliable - less things to go wrong (like dead OLED screen).

You also seem to have genuine 'chip' - at least they did not deface the markings.

The PCB layout looks good - with the micro vias running alongside the input / output traces - but I am not sure if they used the correct substrate.

I bet that FR-4 is what they most likely used - not sure if this is good enough to 6Hz?

I found a link on Rogers .. interesting reading ...

https://www.rogerscorp.com/-/media/project/rogerscorp/documents/articles/english/advanced-connectivity-solutions/circuit-materials-and-high-frequency-losses-of-pcbs.pdf

Its a pity - that we cannot reproduce 'flat' response - obviously there are other 'frequency' factors in play on this device