Feasibility of introducing RF delay via PCB trace?

Question

The Background

My main background is embedded software but I have had to debug the analog circuitry myself so I've picked up a little bit here and there, enough so to describe the problem in a coherent way to someone who could give me a solution. At my work, I was on a project where we were developing a hand-held radar transceiver that operated in the single Ghz frequency range. One particularly thorny issue that we never resolved was the antenna connectors we used introduced ringing artifacts in the receive signal. To prove this as a source we would put 50ohm terminators instead of the antennas and would observe a significant decrease in the noise in the signal. We also observed decreasing noise as we introduced progressively longer lengths of cable between the Tx/Rx connector and their respective antennas. So the initial conclusion was to introduce delay between the connectors and their antennas to dissipate the ringing and make our lives easier on the Rx side.

The Problem

So the real problem is how do we implement that delay, preferable 15-20ns worth, where space is our main constraint (the last board we built was two dual-side 10cm x 10cm boards, one for the electronics and the other for the antennas with the aforementioned connectors between the two boards)? I had once suggested drawing out trace paths between the connectors and antennas in the same manner as illustrated in this question except that's for digital signals and my limited experience with analog signals has taught me that analog RF signals are a lot more pickier than digital signals. Never did get to try out my idea and now the project is on hold for lack of funding.

The Question

So if nothing else I'd like an answer to whether using PCB traces for large analog RF delay is a feasible idea and what pitfalls would need to be considered. Is this even a common practice or does the RF world rely on special circuits to do that function? Another project which I was involved in had the same problem and used a gated switch as a solution which worked but added quite a bit more components (guesstimate of 25% extra space required). I'm mostly looking for a general argument about how this problem could be approached but references would also be appreciated (even if its way above my head!). Thanks.

Answer 1

Creating Delay

To create a delay you want to pick a filter with a large group delay at your frequency range.

Microwaves101 explains it to some extent.

Why a longer line helps

If you are using a longer line and dropping your noise floor you are actually creating a very basic isolator. Now the microwaves101 explanation of an isolator is a tied to an easy to implement microwaves component, the circulator. Now this is a relativly complicated device to think about implementing if you are used to frequencies that Microwave engineers consider DC(like 500kHz). The device that approximates an isolator for you is a diode. If you connect power to a line with a diode that is idea, any reflections/ringing/feedback will be blocked and dissipated by the diode.

What the Cable Does.

As you increase the cable length, you increase cable loss. Lets say you have a number like 1dB loss (which is a gain of -1dB). If reflections are your primary source of noise then a reflection off of your load will travel back to antenna and then back to the load again. This means just from the cable you have caused a 3dB loss(half power). This neglects the fact that the reflection is a rather large loss on both interfaces also. This "isolates" in the respect that as the line gets longer your reflected signals have a larger and larger loss, at-least 3 times what your signal feels.

How Do I Fix It?

This is probably a sign that your load, whatever you are measuring with, is miss-matched, A decent match, giving a reflection of around -20dB coupled with the loss hitting the antenna again should leave your reflections relatively small.

You can fix this with a tuning network, which adds discrete components nearby to attempt to create the correct impedance. You can correct this by fixing any flaws in your layout also. Probably both (based on experience with my own boards).

Another way to approach this is to really cheat with your line connecting the boards. If you are operating over a small frequency range(ie. Able to treat the signal as a single frequency), then you can use stub tuners(which have been mentioned in this thread, just not by name) where the length of the stub acts as a capacitor or inductor(this is very frequency and fab dependent). You can also have a lot of fun by using a length of cable that is a quater-wavelength long. This means that when a signal is reflected, it will return to the antenna with a 90 degree phase shift, then return to the load completing a 180 degree phase shift. This cancels itself out in stead state(not perfectly, but it gets the job done).

Summing it Up

At least you jumped in head first. There are a number of other things that can cause a miss-match, with more application specifics they can be looked into also. For example, a COAX cable causes a missmatch due to its shield having currents on the outside and inside that do not sum to the inner conductors current. I hope this has helped some, and I hope microwaves101 can be of some help. It is by far not the perfect learning resource, but someone is trying.

Answer 2

the reflections and ringing your seeing are a result of impedance mismatches caused by the connectors and stray elements in the design. Running something in the multi Ghz range through connectors and the like isn't overly simple, especially analog stuff.

In the Ghz range stray inductances and capacitances in the PCB make up and trace routing become very significant. There are tools to analyze these situations but they are not cheap, your best bet is to get a book on microwave design that covers designing matching circuits. (sorry i don't have a good recommendation for you)

You can use serpentine traces to delay these types of signals but you have to include any structure you create in your matching analysis as serpentine patterns will create additional stray inductance.

Answer 3

What you're proposing is a common practice. People use serpentine traces to delay signals, though I don't personally know of a case in the 1 GHz range. I've seen estimates before of delays using approximate 1 in^2 for a 7 ns delay, so you'd need to dedicate 2-3 in^2 of board per signal.

Answer 4

The general idea (theoretically) is to put in a stub some distance away from the antenna feed. The length of that stub and how far back it should go is somewhat in the realm of black magic. The tool used to figure this stuff out is called a Smith Chart...