A basic microwave FMCW radar

[LukeW]

I mentioned this in my other thread about choosing high-frequency DC blocking caps, but I thought I would spin it off into a separate thread.

I was inspired a little bit by Greg Charvat's MIT course stuff, but I wanted to go with something a little more ambitious than just screwing SMA connectors onto connectorised off-the-shelf building blocks. And to try teaching myself high-frequency layout.

Also, I wanted to keep the cost down as much as possible, and use components that were easy for me to source - and also create design materials that were suitable for hobbyists or students to re-use around the world at relatively low cost, using components that were accessible in small volumes. In my experience, here in Australia, it is difficult to access most components from Mini-Circuits for example, in small volumes compatible with low-budget hobbyist experiments. A nice goal, for example, would be to entirely have a BOM that can be sourced from the big distributors like Digi-Key so it's all easily accessible.

So, here are some photos of the early work I have come up with so far. It's a work in progress.

A center frequency of about 5.9 GHz was chosen, because the Hittite HMC431 and the HMC717 are the first LNA and VCO pair I could find to match the same frequency range I could find that were both stocked at Digi-Key. This also keeps the antenna size down a bit, and also matches Tony KC6QHP's "Radarduino" project.

I think the power level should stay within the LIPD class. Will check before firing it up

I'm going to be using FR-4, sacrificing a little bit of RF loss at microwave frequencies to gain a substantial advantage in the price and availability of PCB manufacturing capacity, compared to the alternative of using something specialized like PTFE or Rogers fancy dielectrics. How much dielectric loss will there be in the FR4 at 5.8-6 GHz? But obviously FR-4 is very cheap and every cheap PCB fab in the world stocks the stuff, so there is an obvious advantage there over the more exotic RF substrates if you can accept a bit of loss.

Is it cheaper to use FR-4 and deal with the loss, or to fab the board out of unobtainium?

High-ish frequency 0603 chip inductors (3.3nH, Kemet L0603C3N3SRMST) were used in the DC bias path to the gain blocks, along with an appropriate resistor and some caps. Are those inductors suitable at this frequency? In one of the next prototype board revisions, pictured, I replaced the choke in the DC path with a thin meandered l/4 line in place of the 0603 chip inductor. This is really just an experiment so I can see how it behaves, see if it works.

On this board, all the solder mask is deliberately left off along the RF signal path between the VCO, gain blocks, transmit and receive antennas, and mixer. This makes it easier to show the signal path and explain how a radar works to an audience. It also probably makes the microstrip calculation model slightly more accurate because the thin layer of solder mask probably introduces a tiny error in that.

The original layout uses a Mini-Circuits MCA1-85 mixer, and a GP2X+ splitter (not yet placed on the board). However, it is difficult for me to get any stock of the GP2X here, so I have tried replacing it with a Wilkinson divider. 100 ohm resistors are much easier and cheaper to source! Not sure if it will work, but it's a learning exercise, an experiment in black magic. In the next revision the mixer will be replaced with a Hittite HMC219, completely eliminating all the Mini-Circuits components and moving entirely to Hittite parts in the RF chain.

This board basically only consists of the RF stage - the filtering and processing of the IF output from the mixer, as well as the generation of the modulation voltage input to the VCO, is done on another external board which isn't finished yet.

Anyway, that's where things are at so far. I haven't quite gotten around to testing anything in a complete state, and it probably won't work on the first board. Or the first few boards. But it's an interesting educational project.

I don't have any access to a VNA, spectrum analyser for these frequencies or any instrumentation beyond my 100 MHz scope, so in terms of testing things it might just be a matter of pointing it at a neighbour's 5.8 GHz WiFi AP retroreflector target at a certain range, with the VCO sweeping, and see what sort of IF signal comes back and see what happens.

[marshallh]

Keep us updated. I bet you can find someone here with a 6ghz+ specan willing to try some things out if you mail them a demo board.

[w2aew]

Keep us updated. I bet you can find someone here with a 6ghz+ specan willing to try some things out if you mail them a demo board.

Hmmm - I've got access to tools like that 

[G0HZU]

I can offer some advice about the PCB if that helps...
It's hard to see much detail in the images but the PCB layout does look a bit dodgy for 6GHz. It looks like it is a GCPW layout using FR4 but I'm used to seeing a lot of via holes with this approach. Especially under the active devices. How thick is the PCB?

At work, I'm used to having the luxury of Rogers 4003 or 4350 for stuff like this and normally this would mean a 0.02" layer of Rogers on top of some backing FR4 layers. Either way, to maintain RF performance I would expect to see a tightly packed array of plugged grounding vias under the mixer, the VCO and the MMIC amps. Also I'd expect to see lots of vias along the edges of the GCPW sections.

Your layout doesn't have this and my concern would be that the MMIC amps could be unstable (due to unwanted feedback modes in the poorly stitched GCPW) and the mixer might not work to full spec.

Also, I think the 474 amp isn't an ideal choice here as it will struggle to produce enough drive level for the level 7 mixer after the (-3dB) Wilko splitter. The Wilko splitter looks to be very long/large for 6GHz. Am I missing something here? I'd expect it to be made with quarter wave sections at 6GHz. The meander lines look very long on your bias feeds as well?

In terms of test gear you could try and order a Hittite HMC611 log amp eval PCB and use it as a crude form of 'power' detector/meter at 6GHz. Also make some crude but adequate 10dB attenuators for use at 6GHz with SMD resistors and SMA connectors.

If this first PCB turns out to be no good then maybe you could hacksaw out a little square from it containing the VCO and mixer and use it as a crude downconverter to use your scope as an IF detector/monitor when testing the next version of the PCB. i.e. it might work well enough as a very crude signal detector. But the HMC611 logamp would be a better choice.

[LukeW]

OK, so heaps more ground via density on the next board revision. I will put a very close row of dense vias along the edge of the CPWG lines, as per the Hittite reference boards.

The board is 2-layer 1.6mm FR4 with bottom groundplane.

Would it be beneficial to go to a thinner 2-layer FR-4, or 4-layer, and change the CPWG width accordingly?

The divider lines, and the DC bias lines to the 474s, are supposed to be quarter-wavelength.
I just re-calculated it and at 5.9 GHz it should be 6.2 mm (c / (4 * 5.9 GHz * sqrt(4.2)))
That's weird, because I'm sure I calculated it in the first place, but obviously it's not right it is way too big on the existing board for some reason. So it will need to changed on the next revision.

7 dB LO into the mixer + 3dB splitter - 2 dB VCO output = 8dB gain needed in the gain stage, plus a bit extra for board loss. Right??

[marshallh]

oshpark 4layer is "supposedly" FR408 which is somewhat less lossy.

[KJDS]

The output P1dB compression point of a HMC474 at 6GHz is +6dBm. That isn't enough to drive a level 7 mixer even if there's nothing in the way.

[LukeW]

Perhaps the 474 should be changed to a 476. What do you think?

[G0HZU]

7 dB LO into the mixer + 3dB splitter - 2 dB VCO output = 8dB gain needed in the gain stage, plus a bit extra for board loss. Right??

You won't get a reliably adequate drive level for the mixer with the 474.

If the VCO produces +2dBm and this is fed direct to the 474 (as per your current layout) then the 474 will be driven hard to saturation. It will limit out at about +8dBm.

The -3dB Wilko splitter is really going to have about 4dB loss (at a guess) so that means the mixer drive level drops to +4dBm. Most level 7 mixers need at least +3dBm drive level to work well. Below about +2dBm drive level the performance begins to fall off a cliff in terms of conversion loss and linearity and port match etc.

But you will also have some signal loss in your FR4 GCPW between the sections so if you allow another 1.5dB loss for this then you are now right on the margins for your mixer to 'work'.
But it can be worse than this because you also have to factor in the mismatch uncertainty between RF sections. This is because none of the devices have a perfect 50R match at 6GHz. Especially at the mixer LO port. So you could easily accumulate an extra +/-1.5dB drive level uncertainty here and if you are unlucky with the interstage matching then the drive level could dip enough to make the mixer performance drop a LOT.

You definitely need to change the way you drive the mixer

[G0HZU]

Perhaps the 474 should be changed to a 476. What do you think?

The 476 will produce a bit more drive level but I don't know what your performance requirements are for the system. For example, do you need to maintain high reverse isolation from the mixer to the VCO to prevent VCO pulling effects? You don't have much reverse isolation at the moment...

The board is 2-layer 1.6mm FR4 with bottom groundplane.
Would it be beneficial to go to a thinner 2-layer FR-4, or 4-layer, and change the CPWG width accordingly?

You can use FR4 at 6GHz but nearly all my experience is with Rogers 4003/4350 etc.
If I had to use FR4 I would probably use 0.02" or 0.032" thick FR4 and use GCPW with fairly tight ground coupling on the top layer to control the impedance. Then add a backing layer to stiffen the whole PCB.

Also, there is a rough rule of thumb that your GCPW edge stitching vias need to be <=1/8 wavelength apart. You also need to place lots of ground vias under the active devices to preserve performance and RF stability and (in some cases) thermal stability.

[LukeW]

Suppose I attenuate 4dB off the VCO output (which is +2dBm) before the gain block to address the problem of driving it too hard, and change the 474s to 476s. The gain is about 11dB for the 476 (25C at 6 GHz), so with -2dBm in that gets us to +9dBm, this keeps us staying short of the P1dB point which is 10 dBm. Or possibly change the VCO to remove the internal amp?

Then we take off 3dB in the Wilko, so we're hitting +6dBm going into the mixer. This is a lot closer to +7dBm than we would otherwise be. More realistically we might be looking at 4 to 5dBm, if we take off 1-2 dB for extra losses? But still in a workable region.

Performance requirements? Few. Something really simple that basically just works, and then iterative improvement, education and experimentation built on top of that. Really just a PCB-based version of the very simple radar build given in Greg Charvat's MIT courseware.

[G0HZU]

Putting a small attenuator pad after the VCO would help with reverse isolation but I wouldn't worry too much about hitting the MMIC amps with extra drive level. This is a known and accepted method of getting a consistent mixer drive level for some applications. i.e. you drive the MMIC up beyond P1dB and close to the saturation point. This shouldn't harm the MMIC and the VCO throws out high harmonic levels anyway. The datasheet says the 476 has a +5dBm damage level for the drive level if run from just 3V.

I'm assuming that optimum conversion gain and linearity aren't key requirements for your mixer here so you don't 'need' to hit +7dBm mixer drive level. But I agree that you need to aim for a reliable +4dBm LO level at least.

Can you change the board layout to get a more direct (shorter) path for VCO >>pad>>476>>splitter>>mixer? This would reduce the GCPW losses and give you extra margin on the drive level.

[G0HZU]

In the next revision the mixer will be replaced with a Hittite HMC219, completely eliminating all the Mini-Circuits components and moving entirely to Hittite parts in the RF chain.

That's a level 13 mixer. This will need at least +10dBm drive level, probably +11dBm. Do you really need the performance of a level 13 mixer here?

By stark contrast, the Minicircuits MCA1-85 level 7 mixer you have currently chosen only needs about a +4dBm drive level to function very well if you look at the datasheet.

BTW I had a rummage in my Hittite eval board stash here and I can show you an image of the HMC611 eval PCB. This would make a useful signal detector at 6GHz if you could buy one or fit one to a PCB. I get sent all kinds of eval boards when new devices get released. Some boards are really useful  so I keep them after the eval process for future use.

You can also see the way the PCB has a very thin white top layer (maybe 0.01" thick?) for the RF and the rest is made from FR4 for backing.

Also note how many ground vias are placed under the device. You can see the patterns on the reverse side of the PCB. 

[Dago]

Hey guys! Another radar buff checking in.

I've made a simple doppler radar with an YIG oscillator: http://www.dgkelectronics.com/9-ghz-doppler-radar/

And I've been designing an FMCW radar for a long time (on and off): http://www.dgkelectronics.com/new-project-fmcw-radar-made-from-scratch/ This post is old and pretty much everything has changed so...

Currently I have been designing a ~10GHz one with a swept YIG (unlocked at first) but the design part sort of got out of hand (started designing a USB->FPGA->2-channel ADC DAQ thingie for the receiver).

[LukeW]

The rationale for choosing the HMC219 to move away from MiniCircuits mixers is that most MiniCircuits parts are difficult to get here whilst most of the Hittite stuff is DigiKey-stocked, and it would be nice to be able to openly share a project where all the components are relatively easy to get in small quantities from somewhere like Digi-Key.

There's no technical reason why I'm opposed to the MiniCircuits ones, or in favor of the Hittite one.

I might put another attenuator and another gain stage between the Wilko and the mixer to provide higher drive levels needed for the HMC mixer

Perhaps I'll put an attenuator in the layout after the VCO, just in case, so there is space on the PCB to experiment with it if needed, even if it ends up just being 0 dB.

[LukeW]

Re-did layout to make the RF signal path shorter and simpler for less loss.
100pF caps changed to 0402 packages.

Added better ground via "fence" added around RF signal path CPWG, and ground vias under chips.

Gain blocks will be changed to 476s. (Although there is no PCB layout change in this regard. The bias resistor values will also be changed accordingly.)

Attenuator added after the VCO, extra gain block added between the divider and the mixer IF input, along with an attenuator before the mixer. These attenuators can be populated with 0dB, or "played with" to get acceptable levels without needing to re-spin the board. This provides the boosted IF drive level for the Hittite mixer. (Attenuator footprints are single-chip Sumusu pi pads.)

Quarter-wavelength lines in the Wilkinson and the bias tees have had their length corrected.

What do you think?

[KJDS]

That does look a lot more like an RF board.

The obvious things is that I'd want some more stitching vias across the expanses of groundplane.

[Dago]

I would also use a coaxial connector for the IF. It is a relatively "delicate" signal especially before IF amplifiers so shielding it from interference is a good idea.

[calvinlaw]

Great ! Please keep me posted.

BTW, do you have any plan on the choices of the antenna ? and extend the frequency to, i.e., 8-10GHz ?

[KJDS]

The other thing I've noticed is that you may need a small attenuator on the mixer output.

Not only will you get the desired LO-RF at the IF port, you'll also get LO+RF and a 3dB pad prevents this being reflected back into the mixer and causing peculiar affects.

[LukeW]

For antennas I'm thinking about a pair of 5-18 GHz Vivaldi antennas from http://www.wa5vjb.com/

As far as increasing the frequency is concerned, I don't think I will because I think that's adding disadvantages I don't want such as increased cost, decreased availability of components, potentially problems with the FR4 board, and other RF design challenges that are more demanding at high frequencies.

I will make some tweaks and fixes and push everything out as open source eventually, just need to clean up and sort everything out in some reasonable state.

[Dago]

Vivaldis should work great. I made one test vivaldi on FR4 and it seemed to work great. Measured a return loss of -25dB at around 9 GHz (with a directional coupler so this wasn't super exact).



I didn't have two so I didn't measure the coupling between two Vivaldis. I did some measurements on two 12dBi H-plane sectoral horns:



But I can't remember the measured couplings other than "back-to-back" it was "much more" than when the narrow sides were parallel (which is logical).

[LukeW]

Does anyone know where DIY artwork or Gerber files etc for Vivaldis like that might be available?

[LukeW]

Spun up a new board with some tweaks and fixes.

Lots of via stitching, better via fencing density.
IF output comes out to an SMA with a pad between that and the mixer output.
There is also another gain stage and a pad between the splitter and the mixer to allow the LO level to be tweaked if needed. There is also a pad after the VCO before the first gain block.

[LukeW]

OK, so, I have spun up a new board as shown in a couple of pics above, but I'm just checking the datasheets and design before powering up and blowing up some MMICs.

We get +2Bm out of the VCO, and then attenuate 3dB off, so we get -1dBm which is the input to the first HMC476.
The gain of the 476 is 10.5dB (at 6 GHz) and the P1 point is 11dBm, so -1dBm is a nice drive level for the MMIC, keeping it a bit shy of the P1 point.

So we get +9.5dBm out into the Wilkinson, so we get +6.5 dBm out of each side of the splitter. It's probably a little bit less than that due to some loss, but just the sake of argument let's pretend that it's perfectly -3dB each side.

Now, on the current design iteration there is another '476 after the Wilkinson on each side, receiving this +6.5dBm in with only a coupling cap in between, which is obviously no good because we're overdriving the hell out of them, and I'm not going to power them up in this condition. (The maximum input damage threshold is +5dBm). I suppose on the next revision I will attenuate 6dB coming out of each side of the splitter before going into each of the gain blocks. So we will have +0.5dBm coming out of each branch after the Wilkinson and the attenuators, providing a sensible level into the gain blocks.

With a nominal 6 GHz gain of 10.5dB we will get +11dBm out to the transmit antenna and to the mixer LO port. This is a sufficient level, hopefully, both in terms of a reasonable transmission power and a drive level for the HMC219 mixer. This is running both the output HMC476s right up to their P1 point, which is not ideal, but hopefully this works acceptably. We will be running a little bit shy of +11dBm out because of real-world losses on the board, for example we might only be getting say 10dBm out into the mixer - but hopefully this mixer drive level is acceptable. With the extra attenuator installed before the mixer drive amp, the existing attenuator after the amp, at the IF port of the mixer, will be removed.

Anyway, any peer-review or feedback or comments on the above? Thanks!