The 2018/2019 TechNight Radio Project

Below is ongoing updated information about the radio project.

At the end of the design info is RESULTS from building and testing sections of the design.

First, the latest block diagram:


Here's the latest Schematic:


Schematic Notes:

The schematic and PC board are done with EasyEDA software, my favorite free eCAD system.

Version 0.2:
First, I renumbered all the parts, so you can't talk about "R27" in vers 0.1 and have it be the same part in Ver 0.2. I may do that again before I do the board, but then of course I have to freeze it so it stays matching the designators that will be printed on the PC board.

Also in this Ver 0.2, I got rid of the AD603 IF amp and went with a discrete string of gain. This is more flexible and educational, and will run nicely on 5V. The AGC is just brute force PIN diodes to mismatch the stages to lose gain.

Termination on the filter is lossy, but good resistive termination for the filter, which needs a good matched load to keep it's flatness. The gain string is kind of interesting. First is a Cascode pair, which has good reverse isolation, so that the PIN mismatch doesn't get back to the filter and mess it up, and because Cascode stages are cool and have several advantages, like sharing the power supply current with two stages that are in series DC wise.

Next is a pair of stages that are also in DC series, but are just cascaded common-emitter stages otherwise. Also just cool, but you can only get away with this at the beginning of the chain where signals are small and don't run into the low rails of these two stages (like 2V each). Then a more classic common emitter gain stage, before going into the BFO mixer. There are also tuned circuits distributed throughout this gain string, to narrow the bandwidth which reduces total noise power (each stage creates its own broadband noise), and to help stability by limiting the gain to only 10.7 MHz.

The AGC source is very simple and probably won't work like this, but it rectifies the IF signal at the end of the chain, and produces a positive DC level relative to the signal level, and will drive the 6 PIN diodes controlling the radio gain if the signal gets too high. This stops the radio from saturating with strong signals, but reduces the gain so that weak signals are lost if they are close to a strong one (if both are inside the crystal filter bandwidth). We'll see how well it works.

After the mixer, the "audio" signal will be as wide as the crystal filter is, and I intend to use a 30 KHz wide filter in mine, because I happen to have a real good one like that. So I put an audio filter that rolls off at 40 KHz after the mixer, kind of an "anti-aliasing" filter for the next stage, which is the audio digitizer plugged into the Raspberry Pi. If we get a good one it will handle 40 KHz and I can look at that much spectrum at once. BUT, any strong signal in that bandwidth will bother the weak ones, so you might want a narrower filter. Bob's friend Sandie's dog knows the name of this filter: "roof rooof!". Now we'll see if anyone read this far.

Version 0.3:
Most of the changes from last time are clean-up and changes to make the board easier to lay out. Like note the wire crossover at U18 and T3. Doing this on the schematic un-crossed the traces on the board. There are also a lot of bypass capacitors added, note the pile of them at the top. Also, the power supply is split in 3: +5V, +5VR, and +5VT. Oscillators and DDS's run all the time, even though only one is used in TX. Would have to reload it if I powered it down. Might put it to sleep though. Or it can be turned off in software.

Also, I renumbered it again, since it was getting messy with all the changes. I gotta quit doing that, especially now with the board laid out, as renumbering in EasyEDA can mess up the board. So C81 isn't C81 anymore.

I also want to point out how nice it has been using EasyEDA. I'm sold on it. They released a new version just last week with some new additions and fixes. It's not perfect for sure, but completely usable. Amazing to me that it is free, it's better than most of the good stuff I've used over the years. And the most intuitive CAD system I've ever used.
If you want to learn it, you should talk to me first. There are about 5 things I would tell you that would save you lots of startup time. Mostly it is the misuse of words due to translation to English. This is really confusing. Like a "Library" is not a library, it is a part in the library. If you hit "New Library", it brings up the part editor.

Version 0.4:
Lots of changes here for the board layout. Bypass caps added, pinouts changed for easier layout (like P1 connector pin assignments, all changed for direct board traces). Pull-down resistors were added to all the digital inputs, bad to have them floating if not connected. No real electronic design changes to the radio.


Here's the final PC board layout:


PC Board Notes:
This is the final board as it was ordered. This image is actually NOT from the layout software, but from a Gerber file viewer, that looks at the final artwork files you send to the board manufacturer. I made the top layer green and the silkscreen white so it looks almost exactly like the boards we're going to get.
When the board started, it was a blank rectangle with about 300 parts piled up on the side, had to move each part one at a time, then get the layout worked out, looking back and forth between board and schematic. I'm very happy with how it turned out, almost completely single sided, despite a 4-layer board. A few traces on the back side, but none with RF on them. This is possible due to power and ground NOT being on the top or bottom, so still need 4 layers. Note this top side is filled with copper in any open space. There are small via holes all over this board not shown here, connecting ground from top to 2nd layer which is solid ground, and through to the ground fill on the bottom layer. Grounding on an RF board is important, and this will have lots of it.

The front end is in upper left, then going right across the top, mixer, quad amp, filter (upper right). Down the right side is the IF strip, then bottom right is the BFO mixer, then a wideband audio amp and filter. Right now the audio output is 2 pads to solder wires to. The transmitter is in the lower left. The crystal osc reference and DDS chips are in the middle, with data connections on back side over to the main connector on the left. Not such a simple radio, with 271 parts.

The board size is 7" wide by 3.6" vertical, just picked this somewhat randomly, adjusting a little as I went. Seems to be a good size for this prototype, but a final version should be smaller. At that point we'll think about a box that we can make it fit into.


Here's the latest Gain Distribution diagram (nothing new):


For the gain section, green cells are inputs entered. Each row represents a different input level to the radio, and a different IF Gain setting. Red cells are signals that are too large (greater than the max listed for that stage). This represents an overload condition. Yellow is a signal that is probably too small, not enough for the ADC to get much info from.

Note that for a -10 dBm input, regardless of AGC or IF Gain, there is an overload at the first mixer. The radio could be built with a higher power mixer and LO to handle this. Or a lower gain RF amp, which might hurt weak signal sensitivity.

Below the gain section is now a Noise Figure section. This analyzes how the noise and losses and gains of the path affect the sensitivity of the receiver. Unlike the gain, which flows from left to right, showing the effect of gains and losses on the signal as it flows through the radio, this section starts on the right and works itself to the input of the radio, showing how sensitive the radio is given the current stage settings.

At the meeting I showed this analysis with a 3dB attenuator after the mixer and before the filter. The change I made to the design is to add some gain here, but in a special way, that serves the purpose of the attenuator that was there: To provide a good impedance termination for the mixer and the filter. With the new gain, the noise figure at the input goes down from 9dB to 1.7 dB, a great improvement. And now the AD603 IF amps do not set the noise floor of the radio. However, the radio now saturates at smaller signals. We might have to add AGC to this new IF amp, despite it only having 10dB of gain.



Here's a list of the sections that will be addressed separately:

  1. Ref Source
  2. Multiplier for Ref and DDS
  3. DDS1 1st LO
  4. DDS2 BFO
  5. RF Amp
  6. 1st Mixer
  7. Xtal Filter
  8. IF Gain
  9. 2nd Mixer
  10. 2nd IF LPF
  11. ADC
  12. Audio Out
  13. Rasp Pi Software
  14. TX Mixer
  15. TX Bandpass Filter
  16. TX Gain

We will discuss and design each of these sections at TechNight.

Here's the goals for this radio:

1) Cheap - less than $50 not including enclosure or peripherals like AC power supply, keyboard, and screen. Just a PC board, no box. [going to be closer to $80]
2) Pretty good quality - totally usable, not contest grade, but a level below that.
3) Small and low power, for portable operation, so it can be used at home or out camping. It would be surface mount technology, but you won't have to solder any of that, we'd get someone to solder the small stuff.
4) It would use a Raspberry Pi as its controller, so that it can run FT8 (and all the WSJT modes). No external computer required. We will write the controller software for this radio, the "front panel".
5) Hardware not too complicated, maybe only one band. I am thinking 6 meters, but it will be buildable to other bands from 160 thru 2 meters. One circuit board for everything, with maybe 250mW output from TX, use an external power amp on a separate PCB with TR switch and LPF.
6) Decent power out on TX, maybe 20 Watts (with PA).
7) Maybe it is FT8 and CW only, meaning it does not have to be linear. No SSB (but maybe FM).
8) Useful for local TechNight contests, like maybe a worked-all-attendees, worked all of these radios, or worked all zipcodes of members. Something to get people on and using the radio and having fun with it.

All specs are subject to change as we discuss and price things out.

 

RESULTS

Here are results of building and testing sections of the radio.


First, the front end filter, before the first gain stage. This is a double-tuned critically-coupled circuit, with two resonant coil-capacitor pairs. These filters are hard to design, I think most of them get close and then optimize by trial and error. I built about 12 different coils to try here, and bought an array of capacitors to allow me to vary everything. I destroyed a board by soldering and unsoldering 5-pin coils to try things too many times, but the board will still be used to test other sections.

I'm pretty happy with the end result, and call this part done. I ended up with two coils wound in opposite directions, but otherwise identical (tap at same place). With the insertion loss coming out so low, the thought crossed my mind that this should have been a triple-tuned circuit. Maybe I'll try that on the next board. Here are some spectrum analyzer plots with the results.

First is the bandpass response. This is a sweep from 30 to 80 MHz, with markers at 50.1, 52, and 54 MHz. Insertion loss is right around 1.0dB, with rejection at 30 MHz of 45dB, pretty nice. The coupling cap between the two resonators is a little high here, it is 2.7pF and needs to be more like 2.4pF. This is wider than it needs to be, which hurts the far out rejection some. I tried 2.0, the next lower value I had after 2.7, and it is too narrow a bandwidth, so I will get my hands on some parts in the middle to get it right.


The return loss (RL) into the input of the filter with the output terminated is shown below. This is basically the SWR of the circuit: You feed a signal in and see how much of it is returned back at you. You want nothing to be reflected back in the band of operation, you want it to all go through. But you want it all reflected back at you (the antenna) out of band.

Note the two points where the resonators are tuned are very obvious here, unlike in the bandpass. Typically, a low number here means a minimal loss point of the filter, as the filter is then letting you see the 50 ohms of the termination on the other side, which reflects no power as it is a perfect match. But in the middle of the band between the resonators there is a rise in RL, due to the behaviour of the two coils coupled. It's not too bad here, 10dB, and this will get better with a smaller coupling capacitor which will bring the two resonant points closer together.


I think next I will do the 40 MHz reference and the doubler to 80 MHz to drive the DDS chip, since this has a very similar filter in it, but with the difference of wanting as narrow a bandwidth as possible, and willing to tolerate some loss since the signal gets squared up to a logic signal next. So coils should be close if not identical to the ones on the Front End, with smaller coupling cap.

New PCB

In Dec 2019, I went to build up the DDS section of the board, and realized the pinout on the single-gate ICs is backwards, in 6 out of 7 of the parts. How did I manage this one? The part is built correctly in the CAD library, but being a single gate, the schematic symbol had a power and ground pin added to an Exclusive-OR looking symbol, and which power lead was which was not marked. When I rotated the part on the schematic, that reveresed power and ground, yet I assumed the bottom one was still ground. The part symbol now has markings on it.

This would be a hard one to modify the board for. Power and ground pins go right to internal planes, I'd have to bend pins up and run wires over, and these are tiny parts to be doing that with. It's only $50 to make new boards, and there are a few other changes on the list, so I decided to remake the boards. I'm adding another resonator to the RX input to make it triple tuned, and placing the coils further apart. Also putting parts for a resonant terminator on the output of the main mixer, to give it a termination at 90 MHz (the image), which the quadrature amp will probably not provide, as the hybrids won't work that high. No other changes though.

I should really build and test the quadrature IF amp before I do this, but I'll take my chances just to make the DDS prototyping a lot less painful. I'd pay $50 to get out of 12 tiny board mods and wires. Should have boards before the end of January, and will then build up all the DDS circuitry.


Sometime about 2 months ago I got a nice software control package running for the radio, written in Python and Tkinter on the planned Raspberry Pi. I also built up the new version 2 boards I talk about above, with all the DDS frequency generation and the reference oscillator for the radio, and set it up for control by the Pi, and it all worked! I can listen to all the frequency sources on a shortwave radio with its BFO turned on, and hear how the tuning steps work. Seems to work well. I spent a lot of time trying to get the 40-80MHz doubler working, gave up on the cascode design that I originally had and went to an XOR gate doubler, like the 20-40 MHz one in the design. This is working quite well now, though it is not optimized to get rid of x1 feedthrough, it's only about 40dB down at the DDS input, and I'm sure that's affecting the DDS spurious output. I'll put that on the list to check later.

I also built up the new triple-tuned front end and got it working quite well, really happy with this. However, when I added the big gain part (PGA-103), it was unstable, oscillating at around 300 MHz. Shielding on the coils doesn't seem to help, so I think it's purely mismatch on the amp output at other frequencies. Need to add some loss in there. Also on the worry about it later list.


April 5, 2020 - CoronaVirus Era

I've done a lot of work with the quadrature IF amp. There are several issues here, and I'm not done with it. In March, when the decision to cancel TechNight was made due to the coronavirus, I pretty much stopped working on the radio for a few weeks. I started working 100% from home on Friday the 13th of March, and I had some pressing work things to do. I'm pretty happy to continue to have a paying job in these times, so work got some pretty high priority. Still, since I'm home 100% it's hard to separate work and play.

For a couple days, I played with one of the 2 identical amps in that first IF amplifier. This is really the only critical part left on the radio, the IF amp after the filter is pretty simple and not critical. There are a family of RF FETs that can be used in this grounded-gate design, they are a sorting of parts from of a process that is not easily controlled, which sets the pinch-off voltage. The 2N5484/5/6 have pinchoffs of (0.3-3.0), (0.5-4.0), and (2.0 - 6.0) volts. I'm trying to make this radio run off of 5V (since 5V portable power supplies are so common now, for phone recharging mainly), so that eliminates the 3rd one, unless I sort them for pinch-off. It turns out I have to sort them anyway, since in this quadrature amp, the 2 sides have to be as close as possible to each other or you don't get the cancellation of mismatch that the quadrature buys you (with the advantages of the mixer seeing a good load, and the Xtal filter driven from a good load). The voltages listed above are actually negative, the voltage you have to put on the gate to get the FET to bias linearly, but we are running grounded gate, both RF and DC wise, so really that translates to how far positive the source has to go to get into linear operation. But it can't go higher than the supply rail. So I set up a test circuit to run a bunch of these FETs through and sorted them by bias point in a real circuit. Fortunately, a few were pretty close to each other and I could find some matched sets of each type.

But another parameter that varies with this pinchoff is max drain current, which is (1-5, (4-10), and (8-20) mA. This affects the dynamic range I will get out of this amplifier, and I want to stay inside my gain distribution plan. So I ended up choosing the middle one, the 2N5485 (actually the surface mount version, MMBF5485), and sorting for a matched pair in the middle of the range. This would be a serious problem if this radio went into mass production. The parts would be a pain to sort. I'd have to probably change the design to use a different part. The thought keeps crossing my mind about doing that anyway, but there are pros and cons. One big con of using a modern MMIC amp here is the power consumption. To get the same dynamic range, I would probably have to dump 20-30mA into each of these 2 amps, whereas this FET will do the job with 4 mA. I did measurements on dynamic range and it does indeed meet the gain distribution requirements with only 4mA. I even varied the current up and down by changing the source resistor, and watched what that did to saturation level, as well as supply voltage. I have it set for min current for some good margin on saturation, and 4 mA of operation.

The other problem I'm having besides biasing, is matching to get the gain it should have. I ran LTSpice simulations for all the saturation and power supply voltage and all that above, but after doing an ideal match in the model to 50 ohms. The real parts are not like the model. By trial and error, the real circuit requires significantly different matching. By trial and error, I've found some good matching, but the gain is still low, only about 6 dB. This actually might be enough, but the simulation and other uses of this amp I've seen show 10-12 dB. One thing I like about the simple matching is that it is rather high Q, meaning that this is a nice pre-filter before the crystal filter. But the negative side of that is that I can't tune it with fixed value components. One of them has to be adjustable. Interesting to note that other circuits I've seen with grounded-gate FETs seem to have the output inductor tunable, which varies the match in the right direction on the Smith Chart. It is really easy to add the same size coil I'm using all over this radio, but this adds yet another adjustment to the radio. It adds 2 in this case, making it now like 12 coils to wind and adjust, kind of a pain. Have to think about this more...

All for now. Next I'll try to get these IF amps done, then I need to fix the RF amp instability, then I have to get the 20-40MHz double working, which has a critical output power requirement to drive the main first mixer. Should be enough signal out of the XOR doubler to get this, but the filter might cause losses that will force the addition of an LO buffer before the mixer. This is all if I survive the pandemic. If I don't, I'll probably drop this project...

-Dan