Building a 22 GHz network analyzer for under $1000

[JoeyDeng]

I found a cable fit that really well.
anyway im going to take a nap, see u guys later

[JoeyDeng]

also I found that the dielectric of the coaxial exploded when I solder it, so maybe I should use some silver glue or some low temperature solder.

when it exploded the dielectric constant will definitely change. and this will cause a serious impedance mismatch.

[arlo_g]

Re. T splitters

I think that you'll find that T splitters or other attenuators on the port help source match at the expense of directivity and dynamic range.  You can likely still get adequate _corrected_ directivity with a sufficiently good calibration, but reliance on heavy correction doesn't likely come for free.

[Johnny10]

Place a marker

[JoeyDeng]

Posted by: arlo_g
« on: Yesterday at 09:51:09 pm » Insert Quote
Re. T splitters

I think that you'll find that T splitters or other attenuators on the port help source match at the expense of directivity and dynamic range.  You can likely still get adequate _corrected_ directivity with a sufficiently good calibration, but reliance on heavy correction doesn't likely come for free.

Can u explain why the directivity will decrease plsase?

the main reason for the failure of my previous coaxial balun resistor bridge is that the balun was heavily mismatched, and the input impedance of the balun had heavy weight. If we use a t divider to reduce its weight, the directivity should be much better

[EggertEnjoyer123]

Here's my latest simulation. I haven't really tried optimizing anything here yet.

In the first picture, port 1 is on the top, port 2 is on the bottom, port 3 is on the left, and port 4 is on the right. Port 1 contains the device under test, port 2 contains the reference load (basically a 50 ohm standard), port 3 is the input for the signal, and port 4 is the output of the balun.

The second and third pictures show the "transitions" in the balun. The second picture shows the transition from a differential microstrip with a copper ground plane to a differential microstrip with a lossy ground plane. The third picture shows the differential to single ended conversion. I haven't tried to optimize anything (I only made the thickness of the differential microstrips higher because that led to more common mode loss and more directivity).

Here, everything is ideal and symmetric before the balun, so we should expect the S43 to be zero if the balun were perfect and rejected all common mode. In reality, we are getting around -60 dB. (see 4th picture). The insertion loss (S31) is about 10 dB. Basically, the power from the source oscillator will be attenuated by 10dB when it reaches the DUT. In the 5th picture, we can see that the S43 is around -20 to -30 dB. So if all of the signal on port 3 were reflected, the signal would show up on port 4 with -20 to -30dB attenuation. So in total S43 is -30 to -40 in the case where port 1 is unterminated, and -60dB in the case where port 1 is terminated, so my design achieves 20 dB directivity. Also notice that the match is pretty good for port 1 over a pretty large range.

I think I can achieve a higher directivity by adding more lossy material to the top, and also making the differential pair with lossy material ground plane longer. Also, the low frequency performance can be improved by using ferrites (not shown here). It should be pretty easy to snap a few of them around the differential traces.

[JoeyDeng]

https://www.kechuang.org/t/90896
https://sdr-deluxe.com/forum/3-2-3
http://circuitsonline.net/forum/view/140339

EggertEnjoyer123 check this

pawel   Date: Friday, 03.04.2015, 01:48 | Message # 34
Group: Removed





Algorithm of actions, in my understanding, we assemble RF-Bridge  http://hiqsdr.com/images/9/9c/HiQ_RF_Bridge_Circuit.jpg
changing the nominal values of R 5.8 to 68 Ohm, instead of R6.7 one to 150 Ohm, this works more correctly.
A long line can be done differently, faster, one winding on a ring ~ 1000НН in two wires, 8 - 10 turns. It will work up to 29 MHz, but down to 300 kHz! A
50 Ohm SMD resistor is soldered to the DUT input, the REF input is measuring, VSWR to RX, HF_INPUT to TX.
We launch QUISK VNA, the REF button is yellow, nothing is connected to the measuring input!
Press Cal Remove (reset settings), press Cal Open; close the measuring input and press
Cal Short; connect 50 Ohm to the measuring input and press Cal Load. The Run button is active, let's start measuring!
P.S. I look at filters using this bridge, connecting a 50 Ohm load to the second input.
psps After each calibration operation (appearance of inclined lines), press the same Cal... button again - save the result!


Attachments: 5757745.png (262.9 Kb) · 1555180.jpg (98.1 Kb)


Message edited by pawel - Friday, 03.04.2015, 01:53

[EggertEnjoyer123]

I don't think those designs work at millimeter wave.

I think the design I shared earlier should work up to 40 GHz if you tune everything correctly. It'll probably take a few cut and tries to get everything right though.

[EggertEnjoyer123]

Harmonic mixing might be a solution to another one of my problems (which is that I can't generate enough power for the LO past 20 GHz).

I programmed my device to use harmonic mixing past 13.3 GHz. In theory, a double balanced mixer can be ran at odd harmonics (if you look at the harmonics of a square wave). If you turn on the diodes hard enough, there should be enough 3rd harmonic to mix higher frequencies down.

Based on my oscilloscope capture, it seems like the insertion loss increases by about 13 dB. That doesn't really matter though because the isolation is nowhere near good enough right now to see the effects of that.

Anyways, if I do decide to try higher frequencies, harmonic mixing is probably necessary since getting the 15dBm or so at 40 GHz to drive the mixers is going to be challenging (but 15dBm can easily be generated at 13.3 GHz, even with FR4). The XX1000 chip is also interesting (it is a frequency doubler IC which fits nicely with the LMX2820 frequency range), and could be used to increase the range with harmonic mixing. The 13.3 GHz cutoff could also be moved up since I only see issues with LO power at 18 GHz.

I also need to see if my homemade mixer design works with harmonic mixing too.

Picture 1 shows the difference between non-harmonic mixing and harmonic mixing at 13.3GHz. The signal on the left is the IF at 13.3 GHz (no harmonic mixing). The signal on the right is the IF at a little higher than 13.3 GHz (with 3x harmonic mixing). The signal on the left is about 4.3 times larger, which translates to 12.7 dB.

Picture 2 shows a measurement of a 30dB attenuator with harmonic mixing. Everything works.

Picture 3 shows the isolation. It is impossible to tell where the harmonic mixing starts (unlike with old spectrum analyzers, where you can clearly see from the noise floor where harmonic mixing is used). This is because my setup sucks in terms of shielding, and there is a lot of leakage. I think I will have to redesign everything to make the shielding better, but that'll be for when I make version 2 of the VNA.

Picture 4 shows the difference between a warmed up and cold measurement. Before, there would be 10dB difference at 23 GHz because the LO power is too low to drive the mixer diodes properly. Now, there is 1dB of difference between the warmed up and cold measurement. Also, I untwisted and twisted the cable, so there is probably some error due to that. To measure it, I calibrated the device while it was hot, and then I waited for everything to cool down. I quickly powered it back up and loaded the calibration file, and then I ran another measurement to get the difference (though I probably shouldn't have unscrewed the cable). I think I probably should buy a proper torque wrench someday and not just use my fingers.

[EggertEnjoyer123]

My homemade mixer appears to do way worse when it comes to harmonic mixing.

My theory is that the diodes have high capacitance (0.2pF for the one I'm using). This means that the amount of third harmonic generated by the same input signal is smaller. I think beam lead silicon diodes can be bought with <0.1pF capacitance, and the designers of the commercial mixer probably used better diodes. I will probably have to buy GaAs diodes to compete (like the MA4E1317, which is about 3 bucks each from Mouser).

The commercial mixer achieves an uncalibrated S21 of +/- 2dB in this test setup, which means my mixer is behind by over 15dB at certain frequencies. (In normal mode I can get within 5dB of the commercial mixer).

[JoeyDeng]

Now use EggertEnjoyer123's balnanced bridge design we can reach 20db directivity in 8Ghz and 14db in 20Ghz easily, the most important thing is to find the correct coaxial and correct ferrite. I costomized some good ferrite that works pretty well at high frequency. 

using a large resistance resistor like 1kOhm and a 50Ohm term to sample the reference is wise choise in this design, because there is 11db loss of the bridge ,so theres basically 11db directivity of the reference.

My friends helped me to tear down some keys*ght PNA-Xs' bridges and finally I got to know how to match the coaxial of the unbalanced bridge and makes it work at k band. but the material needed is quite expensive I need time to collect the material and get the bridge fully work. 

[irly01]

this is E5071-20G bi-directional coupler

[irly01]

I made a directional coupler , similar to Henrik's bridge, frequency range from 100kHz to 8.5GHz, for measuring S11.

I set the SMA connector (the DUT port) of the directional coupler was not connected to anything as the reference for normalizing the S11 data.
Then, I connected OPEN/SHORT/LOAD at the SMA connector of the DUT port and found that phenomenon:
1. The original directionality ≥ 20dB in the range of 10MHz~6.5GHz (after averaging OPEN and SHORT data).
2.  DUT port connect OPEN and SHORT cal  have wave line that fluctuate up and down by more than ±5dB? (as shown in the open. jpg  figure and short. jpg figure)

I have tested a directional couplers which is made by third-party manufacturers, and the DUT port of the directional coupler only produces small wave line when I connect it with OPEN cal and SHORT cal

[Alextsu]

I made a directional coupler , similar to Henrik's bridge, frequency range from 100kHz to 8.5GHz, for measuring S11.

I set the SMA connector (the DUT port) of the directional coupler was not connected to anything as the reference for normalizing the S11 data.

From the description of Your test setup, it is hard to tell what's wrong.
Please add a photo of the test setup.

[irly01]

Now,I have reduction the amplitude of wave line, through Cap  replaced by a resistor.（myCoupler.jpg）
but lost the circuit DC isolation function.

Through a small capacitor 100pF in parallel with a large capacitor 4.7uF does not show a significant improvement.

Whether official VNA products(like keysight，R&S) also have this issue (DUT port connect to open cal and short cal).  or this phenomenon has been hidden through factory calibration (S11 OSL calibration).

@Alextsu

[Alextsu]

Through a small capacitor 100pF in parallel with a large capacitor 4.7uF does not show a significant improvement.
@Alextsu

Trying to combine the functions of a DC block & broadband coupler in this way, is not a very good idea, IMHO.
And the simple calibration method that You are trying to apply, would not let You eliminate all sources of errors in your setup.

[irly01]

I did not perform calibration, but was puzzled by this phenomenon (other VNAs or bridges do not have such obvious fluctuations).
I think a good directional coupler, whether connected to OPEN cal or SHORT cal on its DUT port, should not have obvious wave line on its S11 trace.

[EggertEnjoyer123]

this is E5071-20G bi-directional coupler

Would it be possible to get a higher resolution picture?

Also commercial directional couplers use a different design which is way better suited for producing a flat response. The design you have has way more variables (like ferrite, component variations, etc) than the purely EM based designs that directional couplers usually have. The advantage of your design is that it works to DC, while EM based directional couplers usually can't reach even 100 MHz. You should be able to measure the coupling for the Agilent design and you'll probably find out that it is not flat too (though you can get it better by using 01005/0201 size components)

[irly01]

this is