Audio Processor

[stax356]

Here is the front end in two versions, modified according to Albert's recommendations.

[stax356]

I was wondering what scheme to choose for pre-emphasis.
Things should be clear here, but there are also different options. Here is a schematic I saw from a commercial DB Elettronica transmitter.

[stax356]

Here is another similar scheme circulating on the internet. Here I really like that an input filter is added and the gain above 15 KHz is limited.

[stax356]

There is also another scheme that caught my attention. It's a bit more complicated but probably worth the effort. What do you think?
Is there anything better than that?

[Albert H]

The last circuit is one that Mark Weiss and I came up with back in the 90s. If you want to use it for 50µs pre-emphasis, change the 330pF capacitors for 220pF The filters get rid of the worst of the harmonics generated by the clipper, but they also introduce a little bit of a phase shift which could compromise the imaging accuracy at high frequencies. The HF clipper can be a good idea if you're prone to driving the audio really hard, but I'd consider tighter and higher order filters.

The DB Elettronica circuit is close to the one I used for the NRG PRO IV coder. It's simple, and has the right shape. It assumes that you're going to have a tight 15 kHz filter after it, though, because the emphasis will continue upwards until the op-amp runs out of gain!

The shape of your "ideal" pre-emphasis is slightly wrong - the curve has to fall off very steeply at 15 kHz, so that there's NO content by the time you get to the pilot at 19kHz

You need to consider the circuit you're going to use for your 15 kHz filter. My favourite is the active elliptical type, using op-amp gyrator circuits to emulate the inductors that you'd have in a passive filter (like the ones we used in the NRG gear).

You also need to decide on the types of op-amp you're going to use. I prefer the TL072 - very high input impedance and very low noise. However, some people argue in favour of the NE5532 or LM833 type - lower noise and higher output current capability.

As usual, the circuits are all compromise!

[stax356]

Albert,
choosing a 15 KHz low pass filter can be quite an interesting exercise!
First I focused on a gyrator. I have looked at several commercial transmitter circuits and most of them use FDNR filters. I was thinking of using the circuit from the Inovonics David 3 stereo encoder.
I also found another schematic with standard component values. I might try this option as well.

[stax356]

Another interesting possibility are MAX291/MAX292/MAX295/MAX296 - 8th-order, lowpass, switched-capacitor filters.
They are very easy to use, I just need to set the correct clock frequency.
MAX291/MAX295 - Butterworth
MAX292/MAX296 - Bessel
Here is a sample application used in the Ramsey STC-1 Stereo FM Transmitter Companion Kit

[stax356]

I even thought about a standard LC Elliptic filter, similar to the modules offered by TOKO many years ago.
When it comes to a single project, it is not a problem to make even non-standard coils with high precision. I think this is a cheap and elegant solution.

[stax356]

While I was playing around designing an elliptic filter with

https://markimicrowave.com/technical-re ... sign-tool/

I noticed that the program allows us to design even a 20 order filter.
I played around with 20 row butterworth and elliptic filters and noticed that they have a significant group delay.
I think it's an interesting idea to combine 15 KHz filter to act as an audio delay for the look ahead limiter as well.
The Butterworth filter if it is 20 order has a good suppression at 19 Khz. But this filter has no constant group delay and will probably sound bad because of phase distortions.
The only suitable one is the Bessel filter, which gives a constant group delay of 40-50 uS, but it is not sharp and it is not suitable for 15 KHz filter.

It seems impossible to combine the two functions - 15 KHz filter and time delay in one module.
I might still have to use some standard filter for 15 KHz, eg FDNR or switched capacitor. And after that, I should do a time delay.

Albert, I know you like bucket-brigade device (BBD). I thought about this option as well. But I'm not sure how good they are and if they are intended for Hi-Fi. Their parameters are not impressive. I don't know if they are overclocked if things don't get even worse. I have also looked at several circuits with them, usually they are used in combination with a compressor before the BBD chip and an expander after it. I think it's too complicated.

Another way to make a time delay is with ADC/MEMORY/DAC. I looked at several schematics, but it is also too complicated.

There are several classic limiters, for example the Neumann BSB 74 Acceleration Limiter. It uses a fully analog time delay with an LC allpass filter.
I think this would be a good solution for a non-professional radio broadcasting.
Albert, would you give some guidance on how to implement such a time delay for a look ahead limiter?

[Krakatoa]

I was wondering what scheme to choose for pre-emphasis.
Things should be clear here, but there are also different options. Here is a schematic I saw from a commercial DB Elettronica transmitter.

You can simulate the circuit in LtSpice to check if the pre-emphasis works as intended. Simple as adding a de-emphasis network at the end of the circuit (for example 10k in series with 5nF to Gnd and take the out put from the junction of the R and C). The Bode plot should be Flat from 0 to 15 kHz at least. If it is not flat, it is because the gain stage does not have enough gain for the needed upwards slope to compensate for the following de-emphasis.

[Albert H]

Any means of generating a delay is tricky, and generally requires a lot of components. Back in the 70s, I worked on delay compressor / limiters made by Pye - these used L-C allpass filters cascaded, each tuned circuit peaked at a different frequency across the audio band. The problems were huge - there were a couple of dozen inductor / capacitor networks, on a large PCB. The audio bandwidth only went up to 7.5kHz (these were designed for mono AM broadcasting), and they were prone to magnetic coupling to nearby transformers, causing hum! Eack of the inductors was adjustable, and the manufacturers provided a table of frequencies that the stages had to be tuned to. They didn't "pump" at all, and allowed the mod levels to be really pushed on Medium Wave.

I did the maths to design a similar L-C delay line for 20 Hz - 15 kHz, and found that I'd need around 55 L-C combinations per channel, and the response would be close to flat, but not perfect.....

The next approach was to try something similar with active gyrator circuits, but I ran into the same issues - huge numbers of high precision components needed, for a relatively poor (and probably quite noisy) delay line.

I went with a digital delay line, using good quality 16-bit A to D and D to A converters, and computer memory modules for the actual delay, clocking the samples through the memory.... The results were really good, but the ICs were very expensive, and required precision components again.

The last analogue try was with overclocked Panasonic BBD chips. These gave the delay I wanted, a good, wide bandwidth, and a reasonable distortion specification (around 0.2%). These little ICs could be quite noisy, but if they were used at their optimum supply voltage and the bias supply was set accurately, I could get the noise over most of the audio bandwidth to around -78dB, which in the circumstances was a pretty good result. I built a number of limiters this way, and they sounded great - no pumping and no overshoots - the limiting level was absolutely clamped...... Unfortunately, Panasonic discontinued the manufacture of the ICs..... I tried most of the "Second Source" versions of the ICs, but many of them couldn't be overclocked, and all of them had lousy noise and distortion specifications!

Subsequently, some friends of mine wrote some software to do delay-line compression and limiting on Linux computers. It was difficult to find good enough sound cards for the purpose, and it was tricky finding soundcards that were either duplex (allowing input and output from the same card simultaneously) or able to work in pairs on the same computer system bus. Eventually I found a Yamaha card that did the job well, and I've recently ported the software to run on a Raspberry Pi 5, but you have to build an interface to allow connection of the sound card to the R Pi! The results are significantly better than the commercial offerings like StereoTool (only really suitable for Tools) and the others.

I considered building dedicated computing hardware for the purpose, as a 19" rack-cased job, but to achieve all I want to do, I'm just reinventing the David V processor / stereo coder. There's no point! (Also, I preferred the sound of the last of the analogue versions of the "David" (version III) anyway!).

I'm going 'round in circles!

So it's back to a minimalist, real-time processor. I've recently built a few experimental circuits based on the "THAT" series of attenuator ICs. My simplest one used a pair of the THAT4301 for a single-band limiter. It sounded pretty good - better than most of the junk that pirates use, but the damn chip has been discontinued!!!

I'm working on a two or three-band design (with integral clippers for the faster transients), that uses absolutely standard components - the most exotic IC is the LM13700 (or 13600), and it uses a bucketful of op-amps. It's going to have gain reduction bargraph meters for each channel, and probably input and output level bargraphs too. The sidechains will be commoned (ie: mono driven), though I thnk that I'm sold on summing the outputs of individual channel sidechain rectifiers - some musical content could give anomolous results, particularly if there's bass signals with differing phases.... Experiment shows that the separate recifiers , with the derived DC level-dependent signals summed, gives the best stereo separation, and keeps the stereo imaging intact.

More work to go,,,,,,

[Albert H]

By the way - the Maxim switched-capacitor filters are just about reasonable performance, but horribly expensive. They also clip horribly if even very slightly overdriven.

My lowpass filter preference is very close to the gyrator-based circuit you show above - that's one I put in EDN some years ago! You can design for lots of the same high accuracy capacitor throughout (economy of scale) and use 1% (or better) resistors which are really cheap these days.

The elliptic filter gives the best phase response, so helps to keep the stereo imaging right. Katruud designed an LPF using a mixture of Bessel and Chebychev stages which gave good results (I'll see if I have it on a drive here somewhere), but the elliptic filter wins in every parameter (except complexity!).

[stax356]

Thanks Krakatoa,
it's a clever trick. I like it.

Thanks Albert,
I am very glad that there is such an experienced person among us. We can learn so much from you.

I understand that audio delay is not a trivial task.
I think the digital method would give the best results but it requires the most effort and design time, expensive ADC, DAC, memory, etc. which is not a good option for a pirate radio broadcast.

The version with BBD chips, in my opinion, has a quality limitation from the conversion principle itself.

While the purely analog version, in my opinion, is not completely hopeless.
A properly designed LC delay line would give decent results. Also, if single units are made, coils can be made and capacitors selected accurately enough, for example with less than 1% tolerance.
If the coils are toroid wound or shielded they should be sufficiently well protected from interference.
I tend to take the time to play this option out in practice to see what happens.
Would you share any guidance on the calculations?

If this turns out to be a working option, I think it's no problem to switch to gyrators. It would indeed require a lot of OP AMPs and precisely selected resistors and capacitors, but so what?
They are not so expensive, and if they are SMD, they will even be of tolerable size.
If low-noise OP AMPs and low value resistors are used, it will be possible to keep the noise at some low levels even if we connect a circuit with 50-100 OP AMPs. It sounds crazy, but I'm willing to give it a try.

I was looking at this limiter recently
https://www.hum-audio.com/laal-mastering-limiter/

It's completely analog and I think it's made that way. The photo shows numerous chips, precision resistors and trimmers. The manufacturer claim to have achieved 0.2mS delay.

In fact, I think that's where we should start! What is the response time of the fastest limiter? What is the smallest delay time that would work?

[Albert H]

200 µs is just about ideal. The attack time on my current analogue limiting stage can be reduced to 150 µs, so there would be enough delay to eliminate the attack altogether. Looking at the picture above, it looks like a ridiculous amount of effort to go to, however.

I'll look out my simplified analogue delay circuit - it gave reasonable results, but needs more stages for a really "bump-free" delay right across the audio bandwidth. All the circuit does is cascade "all-pass" filters, each centred on a small area of the frequency range. Good luck with that! Most of the current commercial versions of the idea split the audio into ¼-octave segments - that implies a hell of a lot of stages for 20 Hz to 15 kHz, and would have to be duplicated for stereo!

Frankly, the BBD approach was the most cost-effective, but the ICs look close to being Unobtanium these days.

[stax356]

Merry Christmas everyone.

I played with a Texas Instruments tool for designing and simulating active filters, and here's what I got:

Filter Type: Allpass
Filter Response: Bessel
Filter Order: 10
No. of Stages: 5
Max Q: 1.415
Passband Frequency: 15 kHz
Time Delay: 200 µs

[Albert H]

The phase errors in your 5 stage version will completely wreck any stereo imaging! To get a sufficiently flat passband, and a sufficiently small phase shift error, you'll need about 11 times as many stages!

It really isn't the way to go!

[stax356]

Yes Albert, I got that.
I understand that everything has a price to pay and if I want to reduce the phase distortions I will have to add many more op amps to keep the same time delay. It can be seen very clearly in the simulations.
I know that the task does not have a simple and effective solution, but I simply cannot stop thinking in this direction.