I need an adjustable delay line for some experiments I am conducting on 10 MHz oscillators. After searching ebay, mouser (for chips that would form the basis for the device) and elsewhere, I decided to design and build my own. As the title of this post suggests, this design is a "poor man's adjustable delay line", since it is cheap, easy to build, inelegant and not high performance. It requires no power supply and the construction techniques are elementary. The downside is it is imprecise and not suitable for frequencies much higher than 10 MHz (see below). I am sharing this design not because it is sophisticated (it isn't) or represents any advancement in delay line design (it doesn't). Rather, I could not find such a design described in detail anywhere and thought if others want to take the same approach, documenting it would be helpful to them.
The ProblemI am interested in characterizing 10 MHz oscillators in a way that might help hobbyists figure out which to use in a particular application. I chose 10 MHz because it is a commonly used frequency for all kinds of purposes (e.g., synchronizing measurement devices in a lab). Focusing on one frequency allows me to get something accomplished in a reasonable amount of time. If others are interested in other frequencies, hopefully my experience (and mistakes) will help them in their efforts. Also, my intial interest was measuring the differences between two 10 MHz distibution amplifiers and so I needed to characterize 10 MHz oscillators before investigating these amplifiers.
A significant problem I faced was bringing two signals into quadrature in order to measure their summed phase noise. Between two arbitrary signals will be an arbitrary phase shift, so to bring them into quadrature, I needed a delay line that could be adjusted between 0 and 100 ns. Achieving perfect quadrature isn't necessary, since the goal is to get the zero crossings of the two signals far enough apart so that a phase detector has sufficient head room to measure phase noise differences at higher frequencies.
Approach, Design and ImplementationThe idea for the device comes from figure 11 of
Frequency Stability Specification and Measurement: High Frequency and Microwave Signals. Shown in the figure is an abstract diagram where different delay elements can be switched in and out of the path of a signal. Unfortunately, there is no detailed information on how this device is constructed.
So, I deep dived into my memory and remembered that ages ago there was a device that used coaxial cables of suitable lengths to implment the delay elements. Each length represented an incremental delay and when several were switched into the signal path the total delay comprised the sum of their individual delays.
From this recollection it was easy to design such a device. I decided to use 8 DPDT switches that would switch appropriately cut lengths of coax into and out of a signal path. To keep costs reasonable, I chose to use some
DPDT toggle switches and
RG-174 coax. Adding some bnc connectors and an enclosure completed the BOM.
There was risk to this approach, since the toggle switches are designed to carry power not rf. My hope (ultimately satisfied) was they could handle 1 V
p-p at 10 MHz.
Figure 1 shows the wiring diagram for each toggle switch. The signal from the previous state is admitted at the left common stud and exits to the next stage through the right common stud. The upper studs are wired together and the lower studs are connected by a length of coax. When the switch selects the upper gang of studs, the signal travels from the left common stud, to the left upper stud, to the right upper stud and then to the right common stud. Very little delay is created in this state. When the switch selects the lower gang of connectors, the signal travels from the left common stud to the left lower stud, through the length of coax, to the right lower stud and then to the right common stud.
Figure 1 -
The input BNC center pin is connected to the left common stud of the first toggle switch and the right common stud of the last toggle switch is connected to the output BNC center pin.
Figure 2 shows the eight DPDT toggle switches assembled on the face plate of the enclosure, the coax running from each BNC to a common toggle switch stud and coax connecting the output stud of the 4th toggle switch to the input of the 5th toggle switch. The upper left and right studs of the other toggle switches are connected by wire with white insulation. Notice the copper wire running down (roughly) the middle of the enclosure face plate. It is used to solder the exterior braid of each coax to ground. This is an extrememly ugly substructure. It represents the application of what I had on hand to solve the coax grounding problem. This is something that warrants better treatment if I were to build another of these devices.
Figure 2 -
Figure 3 shows the enclosure face plate with the coax delay lengths soldered to the toggle switches and placed into the enclosure. Notice how crampled things are. This is another feature of the design I would change in a subsequent version (see
Lessons Learned below)
Figure 3 -
Figure 4 shows the exterior of the enclosure after assembly. The labeling on the face plate will win no awards for beauty, but it does the job. Also, note that the range of delays runs from 6 to 100 ns, not from 0 to 100 ns. The lower bound delay represents how long a signal takes to traverse the device internals when all coax lengths are switched out of the path. This is a measured value.
I included two (rather than one) 1 ns delay lengths in order to provide better selectively at the fine tuning end of the delay adjustment. This also allowed the use of 8 toggle switches, which presented a more symmetric look to the front panel.
Figure 4 -
As stipulated previously, I used RG-174 coax for the delay elements. The lengths I chose for each element were based on delay measurements I made prior to soldering the coax pieces into place. Table 1 shows the lengths used.
| Delay Value | Length (inches) |
| 1 ns | 11 |
| 2 ns | 22 |
| 3 ns | 33 |
| 6 ns | 57 |
| 12 ns | 101 |
| 25 ns | 180 |
| 50 ns | 380 |
Table 1
There are some odd characteristics displayed in the table. To start, the first 3 lengths are multiples of 11". 11"/ns is an unexpected value for RG-174. There are two types of RG-174 - PE and Foam. The former has a velocity factor of .66, while the latter has a velocity factor of .735. If you do the math the expected length of the first type of RG-174 for 1 ns is 7.8", while the expected length for the second type is 8.7". To obtain an expected length of 11" would require a velocity factor of around .93. This seems way to fast for any kind of coax. However, when these segments are measured for delay, the 11" multiple generates the correct value.
The second thing of note is the lengths after 3 ns are not multiples of 11". 57" is not 6*11" (i.e., 66"); 101" is not 12*11", 180" is not 25*11" and 380" is not 50*11". As shown in Table 2, the measured delays for the indicated 6, 12, 25, and 50 ns selections after the coax was soldered to the toggle switches were 7.2, 12.8, 23.6 and 50.8 ns. If you do the math incorporating the lengths, these all represent a velocity factor of ~.66. So, there seems to be something (unknown to me) that produces a speed up in transit time for the shorter lengths of coax. For transparency, it also should be noted that the 380" of coax used for the 50 ns delay came from a different batch, since the RG-174 I bought came as 50' lengths.
These observations imply that anyone attempting to build an adjustable delay line according to the design presented should take the lengths in Table 1 (especially the shorter lengths) as rough guidance, not as normative.
One other issue deserving mention is how tightly the coax is coiled to fit into the enclosure. Some suggest that coax coils should not be smaller than 12" in diameter (a property not satisfied by the coax in the device - see Figure 3). Others say this guidance is erroneous and it doesn't matter now tightly the coax is coiled. I am not an expert in coax coiling effects, so I will leave it to others to argue this out. I can only say that the tightly coiled coax seems to work in this device.
PerformanceTable 2 shows the actual measured delays for each delay element.
| Nominal Delay | Actual Delay |
| 1 ns (first) | 1 ns |
| 1 ns (second) | 1 ns |
| 2 ns | 2 ns |
| 3 ns | 3 ns |
| 6 ns | 7.2 ns |
| 12 ns | 12.8 ns |
| 25 ns | 23.6 ns |
| 50 ns | 50.8 ns |
Table 2
These measurements were made in the following way. A coax connects a Rigol DG 1022 function generator (sourcing 10 MHz at 1 V
p-p) to the first channel of a Rigol 1104Z scope through a Tee. From the Tee runs another coax to the delay line device input. From the delay line device output, a coax runs to channel 2 of the scope, connecting to it through a second Tee. A 50 Ohm terminator connects to the other end of the second Tee. Measurements were made by using cursors on the scope to measure the delay between the signal arriving at channel 1 and that arriving at channel 2.
To find the delay arising from the feed coaxes connecting the delay device to the scope, I disconnected them from the delay line device and directly connected them together. The measured delay was 10.6 ns. I then reconnected them to the device and measured the delay with all delay element coaxes switched out. The resulting delay was 16.8 ns. The values shown in Table 2 are the result of subtracting 16.8 from the value measured when the indicated element was switched into the signal path.
Parenthetically, the indication on the front piece of the enclosure that the device is capable of producing delays from 6 - 100 ns arises from subtracting 10.6 ns from 16.8 ns (to get the lower bound of 6 ns). The upper bound is actually 101.4 ns (which is the value obtained by adding up all the delay times in Table 2).
One important question is whether the DPDT toggle switches specificallly, or the construction generally is suitable for delaying 10 MHz signals without significant distortion. To investigate this, I connected the input of the delay line device to the Tracking Generator on a Siglent SSA 3021X (hacked to turn it into a 3032X) and the delay line output to the SA input. Before doing this I directly connected the TG to the SA input and turned on normalization.
Figure 5 shows the result of scanning the delay line device from 0 Hz (actually 9KHz) to 3.2 GHz with all of the delay line coax elements switched out. And Figure 6 shows the results with all the delay line coax elements switched in.
Figure 5 -
Figure 6 -
As one would expect, the delay line device is not suitable for frequencies in the hundreds of MHz or GHz range. The question is whether it is suitable for 10 MHz.
Figure 7 shows the result of scanning the delay line device from 9KHz to 20 MHz with all of the delay line coax elements switched out. And Figure 8 shows the results with all the delay line coax elements switched in.
Figure 7 -
Figure 8 -
While not perfectly flat, the spectrum in both cases is probably flat enough for low precision measurements (e.g., low precision phase noise measurements in the 20KHz - 200KHz range). For frequencies above 12 MHz, however, the spectrum becomes sufficiently "wavy" that using the delay line device would not be advisable.
Anyone with constructive arguments criticizing the conclusion that the delay line device is suitable for low precision 10 MHz oscillator measurements is invited to present them. In fact any constructive criticism of the design or implementation is welcome.
Also, if anyone wishes to suggest other ways to measure the suitability of the delay line device for low precision 10 MHz oscillator measurments or suggest another way to induce delays on the order of 0-100 ns into the path of a 10 MHz signal, I would appreciate their input.
Lessons LearnedSeveral lessons were learned during the the design and construction of this project.
Positive lessons1. The primative approach taken by the design to implement an adjustable delay line suitable for low precision 10 MHz oscillator measurments, while it would not win any prizes for engineering, seems to work.
2. Lengths for the various delay values are given, which provides useful information to those interested in building this class of delay line. However, the shorter length values are less reliable than the longer ones. Someone reproducing the design should use these lengths as a guide, not as a specification.
Negative lessons1. The aesthetics of the implementation are not inspiring.
2. The enclosure is too small. I would use a larger enclosure if I had to do the project over again.
3. The grounding "line" needs to be redesigned to provide convenient access by a coax sheathing braid to ground. The way I had to route the sheaths to ground through patches and stub wires is very clumsy.
Next Version (Speculation)While I will probably be busy measuring 10 MHz oscillators for the foreseeable future, I did think a little bit about how I would approach a second version of this type of adjustable delay line.
First, I would research the use of PC Board traces to replace the coax lines. According to
this article, signals propagate at about 6"/ns on PC board traces. So, to implement a 50 ns delay element would require a 300" trace. Of course, such a long trace (and those for the other delay values) would be a transmission line and controling radiation and interference would require the use of special PC board layout techniques. Designing such a delay line is way outside my area of expertise, but I would imagine using a multilayer board with ground planes above and below the delay line trace. Perhaps it would also be necessary to separate any wrapped lengths of the delay trace by ground traces to eliminate cross-talk.
Another improvement would be to use electronic analog RF switches to route the path of the signal. The manual toggle switches then would only select the path using a logic 0 or 1 connected to switch control inputs. This would eliminate the necessity of the toggle switches forwarding RF signals through their internals.
Home
Help
Search
Login
Register
