In this thread I will describe my endeavors with designing my own three band audio processor aimed at AM hobbyists. This is a followup from my previous work on a very simple FM audio processor that was essentially a dual band limiter, low pass filter, and unfiltered clipper output utilizing high frequency pre-emphasis correction. I've learned a heck of a lot on this topic from 2012 when I designed that and I plan to share everything I know here to fellow pirates and part 15 operators along with anyone else who is curious.
I will start this with a primer on audio processing in general and move on to the circuitry involved in solving these problems, and eventually my finished circuit which has come along nicely shared to the general public free to copy and use.
This will be a multiple part thread and eventually converted to PDF. I do not claim to be a writer or professional engineer. This is purely out of the hobbyist spirit to guide others along with myself for learning and entertainment purposes only.
If you're only interested in the schematic please give me some time. I am writing this like a blog and need to start with the basics first so we all are on the same page.
My project isn't even complete yet, but I have enough of it designed now that I feel comfortable giving this post a start.
First will be some background. The second major part will be completion of the schematic with scope/spectrum images and possibly sound samples. Third major block of text and info will be of the final design before I put it in use at my own station
April 10, 2019
A primer on audio processing as Kage knows it from the analog realm and experience with modern digital processing:
I have spent a great deal of time in the not so distant past studying how FM broadcast audio processing chains work along with the mostly all in one solutions that many stations still use. There were a few things that stuck out to me with my previous FM design that I really want to tackle but for AM instead for two reasons..
-AM transmission is relatively simple and requires far less circuitry, we can design monophonic processing whereas stereo transmission often used on FM requires duplicate stages of processing and sound stage platform sharing to reduce "ping-pong" aural tilt.
-With AM we don't need to focus on protecting a stereo pilot, and pre-emphasis can be designed simpler given that our audio cutoff is far less than 15kHz for stereophonic broadcasts on FM.
One of the main things I wished to focus on after other people suggested the idea was multiband audio compression/limiting. This is especially important with AM since loudness is king given all the various things that interfere with the mode of modulation over FM. I've always been a fan of keeping FM audio clean and mostly unprocessed outside of basic brick wall limiting on the stereo sound stage to keep dynamics since this is where FM rivals in sound reproduction, whereas we know AM and especially the lower frequencies used for broadcast need to fight hard with modern electronic noise and lack of "capture effect" rewarded by FM.
There's one huge issue with multiband compression though, it sucks dynamics out and creates listener fatigue when overused. This means there must be even another stage before compression to cut back the pull-up of soft passages that contain tape hiss, record pops, 60hz hum, or other background noise in a studio. This is where downward audio expansion comes in handy which is the exact opposite of compression, often simplified in the cheapest processors as audio gates that work wide band on audio. I chose to go further with my design by expanding on the three audio bands independently much like the well known Dorrough DAP 310 or Texar Audio Prism which I was highly influenced by. The balance between these two major stages, or rather 6 (3 audio bands of operation for expansion to compression) make my project compare to the higher end units of other manufactures while remaining on a hobbyist price tag using common op amps circuits.
It gets worse.. What happens when you feed this multiband compander (expansion/compression) with random volume levels from the studio at the mixer? This causes large variances at how the circuitry will create overall sound presentation since both the compressor and expansion stages being split into various bands will suck up or back down frequency selective ranges causing great variance in program sound stage given nothing more than how far the DJ cranks up the fader so we need to introduce yet another stage!
This is where the *slow* automatic gain control (AGC) comes into play, and is the first stage of an audio processor, so that it keeps the overall program material at a steady level with molasses level changes so the rest of the compression/expansion circuitry can do its thing (while the DJ is drunk and twiddling studio mixer controls) as the later processing stages expect a semi-standard level of input, preferably preceding an adjustable manual "processing" control level which feeds the seemingly constant AGC output to the later stages where loudness can be contoured properly outside of what drunken party happens in the studio.
But wait there's more...
So far we have our studio mixer board running out to our hypothetical rack of AGC -> audio crossover (split audio bands for multiband) -> downward expansion circuits -> compression circuits.
At this point it's safe to combine the compressor outputs now that the bands were split and worked individually upon to increase overall loudness across music/voice spectrum. In my project I split the audio into three ranges, low (25hz - 250hz), medium (250hz - 3.5kHz), and high (3.5kHz - 10+kHz). These crossover points seem to be a good standard for AM broadcasting as it works on music bass, vocal range, and high pitches individually. I found 3dB crossover points were good enough much like the DAP 310 from the 1970s, it's enough to bring classic rock up to volume without audible pumping.
This however is still not enough to be useful for broadcast and clean loudness. Anyone who has worked with broadcast processing knows what comes next is even equally complex and there's a reason why this is done in the digital realm now using software like Stereo Tool..
We now have to add pre-emphasis to brighten up the AM broadcast because modern AM radios have built in de-emphasis. This is a steady rise in frequency from 1kHz to 10kHz, rising 10dB at 10kHz. It's even worse with FM rising to 17dB at 15kHz.
This leads to a serious audio engineering issue V.S. modulation maxima, we have to pass the huge increase of high frequency audio to the transmitter without over modulation and that shrill noise, but how?
If we sweep the audio processor at this point with a sine wave from bass frequencies to the highest pitch frequencies we notice modulation climbs from acceptable to grossly over modulated at the highest frequencies to accommodate for 75uS pre-emphasis (50uS in Europe).
Welcome to the next stage in the air chain, the audio clipper.
This is often done with hard diode clipping or a combination of soft/hard diode clipping, sometimes soft clipping early on before combining the compressor outputs.
This simply sheers off the loud over modulation of increasing the equalization curve with pre-emphasis which works because most music and voice doesn't drive those high frequencies up too high anyways unless you broadcast electronic music, but we need to make sure not to go beyond <0% modulation where the transmitter carrier cuts off, and some known value of positive modulation.
Oh and about that positive modulation, we can use higher positive modulation than negative on AM within reason to increase loudness. Legal licensed radio stations can use up to %125 positive modulation, and some %95 negative modulation before risking carrier pinch off which causes interference to stations next to ours because of channel splatter from the rapid cutoff of carrier.
The clipper stage is where this is usually accomplished by DC unbalancing the audio before heading into diode clipping so that one side of the audio waveform is sheered off sooner than the other on peaks. To follow nice symmetry this all can be preceded by all-pass-filters to average out asymmetrical peeks before clipping. A topic for another time.
Sounds good so far, but we now just created terrible audio harmonics by using diode clipping. Any clipping of audio causes nasty square wave harmonics, even if we don't hear it because it's high frequency content, it now exists because the audio waveform is no longer pure. Running this audio at this point into the AM transmitters modulator will sound fine, no one will notice there's an issue and you'd think we now have what is needed to process our music broadcasting out to the public, it will sound loud and proud and this is exactly what we want, you're now doing everything the big guys do but.... unless you're the listener of a radio station adjacent to yours...
We're not done yet
What we've done with the clipper is widen up our audio bandwidth substantially. We cured one issue with pre-emphasis overshoot but caused widening of broadcast bandwidth far outside our channel causing splatter on stations near us. So... we can filter the clippers output right so we still remain <10kHz audio? Yes we can.. sort of.
We need to filter it, but by doing so we cause what is called audio filter overshoot or "ringing". This will cause the loudest peaks of our clipped pre-emphasized audio to be a few dB louder to the modulator following it compared to our seemingly leveled clipped bricked walled audio. So we just reintroduced the issue again but from a different process because of the nature of audio low pass filtering. You can see this process happen by taking a simple stereo equalizer and lowering the high frequencies and watch as you drive a square wave test tone into it and sweep it from low to high compared to a sine sweep, it will show peaks at high frequencies where you thought you reduced them because of the nature of Fourier transform and natural audio filter ringing.
Finally the last fix, and the last stage of the audio processor, and it's an important one...
Overshoot-compensation..
This is simply one of the most hairy stages of the entire broadcast audio processor. What we need to do is somehow bring our clipped/filtered overshoots back to a nice brick walled window of audio for our modulator even after all of that limiting and AGC and everything else, you'd think at this point you'd get 100% modulation without issue right? You'd be wrong sadly. The simplest way to do this is to phase correct overshoots because we know that overshoot is caused by a phase difference in audio filtering (phase shifts in audio filters as frequency increases), so we can take something like an OP Amp stage and phase correct audio by subtracting overshoot by analog prediction.
This alone gets into territory of patent protections and the bread and butter of high end designs.
I used the simplest idea I could find for this and not sure who owns the idea behind the invention so I won't get too far into this topic other than to say it's an art, and there's a reason why Orban, CRL, Dorrough, and other major broadcast industry companies own the secrets to this stage of their equipment. Even I am not absolute sure I understand the underpinnings but luckily for hobbyists we can look at the patents and recreate our own as long as we don't resell the idea.
I spent the last month trying to find a workable circuit that could remove this filter overshoot to compensate for ringing of high frequencies from the final pre-emphasis stage and what I can say is that I don't know how I made it work, but it does, loosely based off of reading a ton of patents.
A tiny teaser of the work I have been soldering up after rigorous design work for the last few months...
For clarification, that's from top left to bottom right AGC, three band active crossover circuit, three band expansion, 12db user EQ, 3 band compressor/limiter/soft clipping, (moving onto breadboard in reverse) asymmetric clipper with POT to control between 100/100% modulation to +-125%/-+85% modulation, 9 pole 10kHz filter, overshoot compensation circuit.
See you all in the next update.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
