Direct conversion receivers - Curing Audio Instability

You might be forgiven for thinking that after sorting out a bandpass filter and mixer that the hard part is over. Unfortunately, this is the part where art is just as important as science. The two largest issues to address are instability and selectivity.

Instability arises from two causes: feedback via the positive supply rail and feedback via the negative supply rail. The positive supply rail is readily fixed with RC decoupling and voltage regulation. The negative supply feedback is prevented with grounding and layout. Let's look at each audio stage and see what was done.

Audio Amp:

Since I'm running several watts of audio output we have to watch feedback carefully. The circuit board puts several capacitors close to the audio amplifier chip. But that is only the start and active regulation is used elsewhere to reduce feedback via the positive supply rail. Now the subtlety of layout takes over. The power is connected close to the audio amplifier chip. Just like a "star ground", the remainder of the receiver draws power from the far end of the track on which the capacitors sit. Don't think of the track as being 13.8 volts. Instead, consider it as high current. We want to power the remainder of the receiver from a point that has the lowest current. The point I use still has to be decoupled further and I run the rest of the receiver from a low drop out regulator of 12 volts.

The same concept applies to the negative rail. Current in the negative rail is what gives rise to a voltage differential between the point the negative supply connects to the PCB, or local ground, and any other point connected to the local ground via a trace. To minimise the current flowing between local ground and anything other than the audio amplifier ground, I break the ground into two major sections.

The first is just around the audio chip and power connector on the bottom sides of the PCB. The attach image tries to explain this. The green tracks are ground on the bottom layer. The only parts sharing this ground plane are the audio amp, the supply capacitors and the power connection. The top ground plane only connect to this green layer via the power supply pin. This prevents any circulating current from flowing through the ground plane on the top side of the board.

Ground Traces for Audio Amp

With a grounding scheme that hopefully minimises feedback on the negative rail we can look at the other audio stages.

Audio Pre-amp:

The first audio stage uses a discrete bipolar amplifier. In practice I never found an op amp in my junk box to be as quiet as this circuit which I first saw in the August 1992 edition of QST in an article by Rick Campbell. My implementation is a bit casual in comparison. I simply used what I had to hand that was close to Rick's circuit. I have found the active decoupling circuit on the positive rail sufficient when taken from the regulated 12 volt line.Before I used a regulated 12 volts I had instability problems, but that could have been also partly due to layout.

Audio Low Pass Filter:

The output of the audio pre-amp drives a low noise op-amp for further gain before an 8 pole low pass filter. I have used both switch capacitor filters and active filters based on op-amps. I don't have a strong preference either way. Both approaches work really well but I have stuck with the op amp approach this time.

In the past I used a rail splitter to bias the op amps. The drawback is any noise or feedback on the positive and negative rails finds itself at the input connected to the rail splitter. It will be attenuated, but is still present. This time I  took the regulated 12 volts and inverted it to negative 12 volts.  I was hoping for a magic bullet but I'm not sure if this extra complexity made much difference. But since the receiver is working I'll follow the wise saying "If it ain't broke, don't fix it!"

Other Grounding Issues:

One issue I had when building this receiver took some time to understand. When completed it was not as sensitive as I expected. The clue to fixing this arose when I disconnected the local oscillator and the loudspeaker noise increased. This occurred even when I removed the inductor between the audio pre-amp and the product detector.

With the oscilloscope probe grounded at the negative power supply pin I poked around and found the 74HC04 squarer I was using was generating lot's of hash with no local oscillator input. Forcing the input pin low with a screwdriver blade eliminated the noise and hash. As the circuit to be posted later showed there was heavy decoupling on the power supply to this part. Holding a tantalum capacitor across the existing decoupling capacitors next to this IC made no difference. Looking at the circuit board revealed the cause:

As current flowed from the ground pin of the IC it was in part flowing past the grounding points of the pre-amp (Shown with * in the diagram above). Since current flowing through a finite resistance gives rise to a voltage this was injecting noise into the pre-amp via the grounded components. The fix was a via to the top layer at the ground pin and for good measure a via at each group of decoupling capacitors on the positive supply to this part. (Shown with # in the diagram above). The result was an increase of sensitivity of around 15dB!

Summary

Curing audio instability and feedback is a dark art. I am only just starting to get my head around this issue so any useful advice is really appreciated. One thing I have learned is that connecting the ground lead of your cro to the negative supply rail of your power supply where it connects to the pcb then poking around is very illuminating.



Regards
Richard VK6TT

Direct Conversion Receivers - RF Bandpass Filters

Today I'm turning my attention to the RF bandpass filter used between the aerial and the product detector. If you see a circuit that just uses a single tuned circuit for filtering then move on. Really. The maths and software for working out values has existed for so long that there is no excuse for single tuned circuits.

So let's move on to considering double versus triple tuned circuits. It's very easy to use filter software like Elsie, AADE or Iowa Hills. My favourite is the Iowa Hills software. It does 80% of what I want with the least effort so I always start with this software.However, I'll step you through how I made an error and suffered from a deaf receiver till I worked it out.

This first part was done in Elsie but it was not the fault of the software. Rather, it was a case of garbage in, garbage out. Here is a double band pass filter for 40m. Since it is a 300KHz segment we will use that as the bandwidth.

I like it, but perhaps a triple band pass filter will look better?

I'm getting excited. Let's compare the two responses on the same graph to see what we trade-off for the extra selectivity:

Okay, we give up 2dB of loss in the passband for much better selectivity. Sounds good to me. Or is it? At this point I was pulling my hair out with a deaf receiver. How did that happen? Quite apart from using by mistake the wrong value parts when I built the filter, it transpires that the curves above were designed around inductors with an unloaded Q (Qu) of 200. What happens when we use a more realistic Qu of 70 for the inductors? We suffer an extra 5dB of loss in the case of the triple band pass filter with this bandwidth.

Which gives us almost 8dB of loss before the product detector. In urban areas that may not be an issue if you have high noise levels. But in my location 8dB sounds excessive given the low noise levels I enjoy. And this extra loss compounded a problem I had which we will discuss in another post concerning low noise op amps.

I learned my lesson. Check your assumptions before you build. In practice I start with a bandwidth of 10% of the centre frequency, say 715KHz in the case of 40m. I then adjust the bandwidth and terminating impedances until I get some standard capacitor values.

You can iterate away a lot of time if you're not careful. My favourite package, the Iowa Hills Filter Designer package, is well suited to this task. After just a few clicks I get the following circuit and response:

Which all looks really nice. But no loss? That doesn't make sense. Remember to check the real parts box! The resulting loss is 1.5dB. That was so easy I repeat the exercise for a triple bandpass filter. I get a loss of approximately 2.1dB. At the third harmonic of the local oscillator, or 21MHz, the 2 pole filter has an attenuation of 15dB. In contrast, the 3 pole filter has an attenuation of 42dB. I don't want to be hearing stations on 21MHz when I'm tuning around on 7MHz. Given the improved selectivity of the triple band pass filter over the double, I elected to go with the triple despite the slightly higher loss. Finally, I tweak the design by broadening the response while watching the loss at 7MHz and 21Mhz. I end up with less loss than the 2 pole filter but better attenuation at 21MHz.

All of which took me less time to do than write! Here are the two filters compared:

Which is all fine so far. We have standard capacitor values. Inductors are always harder to source. If your junkbox has variable inductors of a nominal 600nH then you're in business. The Q of these is likely to be around 70. Coilcraft make a really nice series of 10mm vairable inductors which would be ideal and with Qu in the mid 70's match what I assumed in the software. (Coilcraft, if you're reading this I'd love some of your stuff but it's so damn expensive in Australia that I can't afford it!)

So off to the junkbox. What? No 600nH inductors.

Lots of recovered variable inductors around 3uH though. Back into the software. Hmnnn. 3uH is too large a value for the shunt LC tank to work. I try the series option just in case. Viola! I get a practical design with losses comparable to the version discussed above and a suggested 90dB of attenuation at 21MHz. I really doubt I can achieve that in practice without serious shielding but it should go a long way towards attenuating signals on harmonics of the local oscillator. Of course, there is no free lunch here. I have less attenuation of signals below 40m now. Of concern is a nearby 50KW AM broadcast signal on 720kHz breaking through but to date I haven't noticed any such problems.

Final 3 Pole 40m Bandpass Filter - Series versus Parallel configuration

I found the triple bandpass filter easy to peak using a signal generator on 7150kHz and a local oscillator or 7149kHz. I tuned for maximum audio after the audio low pass filter, the subject of the next post.

I hope you're inspired to download the Iowa Hills software and start rummaging through your junkbox for variable inductors.

73's
Richard VK6TT

Direct Conversion Receivers - Simplistic designs do not do them justice

I was pulling apart some goodies that I recently acquired to see what was inside and found myself looking at switched capacitor filters, NE570 compander chips and low noise op amps. It took me back to when I was actively building HF transceivers consisting of Double Sideband Transmitters and Direct Conversion (DC) receivers and the long path to success. I found the final result well worthwhile so I thought it was worth updating my receiver onto one PCB and posting about what I learnt.

Firstly, I'm going to be controversial and point out that the majority of DC receiver projects usually found on the web are not serious receivers. I don't mean they aren't enjoyable, rewarding or capable of hearing signals. They can do all of that. But they are just cheap. Thrown together with a minimum of parts as if saving a few pennies becomes a virtue allowing you to overlook the shortcomings.

Nope. I've been at the cheap end and it is very unsatisfying. My design criteria are:

• lots of volume from a speaker since headphones don't work if you're moving about the workshop while you listen to a net,
• no audio instability, 
• no hum, 
• good AGC so you are not jockeying the volume control, 
• good audio filtering, and 
• a squelch. 

In short, I want a DC receiver that can be used without having to apologise for it's failings.

I was able to achieve these goals and build a 40m DSB transmitter / DC receiver combination that I actually used in a contest. This is not suggesting it was competition grade. Rather, it was a radio which worked well enough that it had to be punished to uncover shortcomings. And while it struggled at times with a band of strong contest signals I considered the results acceptable.There were many, many, iterations over more than a decade before I deemed the results acceptable.

Over three or four posts I will take you through the final configuration that achieved this stage be stage. I will discuss what I tried, the shortfalls and the improvements (or corrections!) I made before I was satisfied.

Today we will start with the product detector, or mixer. This is perhaps the single most important design decision that shapes everything else. In chronological order I've tried the following product detectors:

1. Discrete transistor product detector
2. LM1496
3. SBL-1 diode mixers and equivalents
4. FST3125 mixers

What I found was they all worked, but some are better than others. The discrete transistor mixer, used as a product detector, actually worked surprisingly well. I couldn't find a LM1496 back then, or a CA3028 or similar. So I lashed one together from discrete transistors (BC109's from memory) as if it was a CA3028. It was so long ago I couldn't find my notes but it was similar to this one: Discrete transistor Mixer

Eventually I found a LM1496 and that  looked more conventional, see below. The output of the product detector was feeding a low pass filter built around a LM324. As I discovered later the LM324 was too noisy. But since I didn't know any better I was running the LM1496 at maximum gain by connecting a capacitor between pins 2 and 3. I could hear stations and I made several contacts with the transceiver this was used in, but I still suffered from cross modulation and strong signals overloading it.

Lesson Learnt: With no RF preamp you need a low noise audio stage after the product detector even when using an active mixer.

LM1496 Product Detector

So I moved on to a diode double balanced mixer, the SBL1. The holy grail I thought. A triple band pass filter to reduce the chance of cross modulation and overloading from transmitters outside the 40m band.

SBL Product Detector (Note: Filter values shown proved to be wrong!)

But I still had a deaf receiver. About this time my collection of test equipment had grown to include a HP signal generator and an oscilloscope. Now I could make repeatable measurements i.e. a measurement made today would be the same as a measurement made 2 weeks prior. Poking and prodding revealed the RF bandpass filter had too much loss and the audio strip was still too noisy.


The temporary fix was to install a RF preamp between the band pass filter and the mixer. This transformed the radio. Ultimately I sorted out the loss in the RF bandpass filter, fitted a low noise audio amp directly after the mixer and was able to remove the RF preamp.

And while this served as the product detector for a good year or more I still wasn't happy. I live in a low noise location so connecting the aerial did not always produce a large increase in noise in the loudspeaker. So I built a product detector around a FST3125. With my now further improved collection of test gear I could see that this approach had promise. Importantly, this IC is far cheaper and more readily available than the SBL1. It can often be found on early server motherboards for free.

While I was satisfied with the SBL1 style mixer, I consider the FST3125 product detector to be at least as good and far cheaper. Next time I will discuss the RF bandpass filter..

More on that later.
73's
Richard