Getting the Mod right

[Albert H]

There have been a few questions lately about audio processing for broadcast.

There are a few basics that everyone needs to understand:

FM deviation is limited by convention to 75kHz either side of the nominal transmit frequency.

Audio bandwidth is limited to 20Hz to 15kHz. Anything outside that range will cause distortion or other unwanted artefacts.

Pre-emphasis ("treble boost") is applied to the transmitted audio. The pre-emphasis time constant is 50uS for Europe and 75uS for the Americas. Use of the Yank time constant in Europe will lead to toppy audio which will be quieter than the rest.

Deviation limiting is mandatory - even if it's as crude as a couple of diodes across the audio. Over deviation will cause adjacent frequency interference and reduce the average power density of the signal. It will also cause gross distortion in any listener's receiver, so will rapidly lose you audience.

There is a fairly small dynamic range (practically) available in FM broadcasting. In theory, it could be as much as 50 dB, but the reality is that anything below about -20dB below peak deviation will be lost in the noise (especially in a noisy environment like a car). Many better quality CDs will have a dynamic range of up to 35dB, so compression is essential.

Remember - you never want to over-deviate, so a limiter is mandatory. The best kind are delay-line based, but something as crude as the well-known "Pira" circuit can give reasonable results.

Software processing is another option and is commonly used in professional broadcasting. Digital nonsense like this introduces artefacts of its own, so should really be used as a last resort.....

[NOYB]

Well said. I agree with everything said above apart from the last point. I especially like the point about power density, which is easily missed.

Digital does not have to be bad, and can be very good. If you have a spare old PC kicking around, it could be much cheaper than a HW processor. Some software can also do the stereo MPX creation and RDS - so then it starts to look very attractive. Even on a built-in sound card the resultant MPX signal can be near perfect - arguably better than discrete analogue circuitry could achieve.

Analogue or digital though, the trick is to remember "less is more"!

[Albert H]

One point - I've tried dozens of software stereo coders, compressors, limiters and all the rest. In every case - irrespective of the type of sound card used, there have been additional unwanted artefacts which - if fed to a transmitter - would cause additional sub-carriers and certainly make the signal nothing like "clean" enough for serious use. I have used a PC-based airchain for a festival station a few times, but the modulation input to the transmitter had to have substantial low pass filtering to get rid of the crud artefacts!

[pjeva]

I'm writing post about software tools and how to set them to give reasonable results. Both for compression & limiting section and deviation measuring. Many computers have decent enough sound cards for this use, but need some tweaking


Sent from my iPhone using Tapatalk

[Albert H]

I'm writing post about software tools and how to set them to give reasonable results. Both for compression & limiting section and deviation measuring. Many computers have decent enough sound cards for this use, but need some tweaking

Sent from my iPhone using Tapatalk

You're right, but the out-of-band products (that will produce spurs either side of the normal modulation bandwidth) have to be tamed by use of a lowpass filter on the way into the modulator!

[thewisepranker]

I've got a Hewlett-Packard spectrum analyser that goes down to 20 Hz, which I think might even have a 600 Ω input. It's broken at the moment (someone dropped something on the input ports, which snapped them off) but when (if) I fix it I'll measure the output of my 192 kHz sound card, running Breakaway Broadcast with Airomate running the RDS, directly. It weighs about 40 kg, bastard thing.
I also have an Anritsu spectrum analyser that goes down to 9 or 10 kHz if I never get round to fixing the HP.

I genuinely don't know what to expect.

[NOYB]

I've got a Hewlett-Packard spectrum analyser that ...weighs about 40 kg, ..

Is that a HP141 ? - wonderful machine, but a definite 2 man lift! I prefer the 8590 series though - does not need to be manually calibrated every time it's switched on!

[NOYB]

One point - I've tried dozens of software stereo coders ....

How far down are the products you are talking about, and how far away from the carrier? When I tried Breakaway on a bog-standard Dell Celeron laptop it looked clean as a whistle. Maybe I'm not looking hard enough?

[thewisepranker]

It's a 3585a. It only goes up to 40 MHz but it's got a tracking generator and was extremely cheap. I lied - it's got a 1 MΩ input as well as the usual 50, not 600. I knew it had two input ports.

The HP 859x is everywhere, and rightly so - it's a decent analyser. The place I used to work had 4 of them in various flavours - one of them went up to 22 GHz. I wasn't unfortunate enough to have to work up that high, though. I nearly bought an 8595E but got too good a deal on my Anritsu MS2601A, with a few other bits thrown in, to turn it down. This is the best bit, though:

[thewisepranker]

Just a few points on the subject rather than waffling on about gear.
Breakaway is a very decent bit of kit - it lets you set up your maximum deviation limit using the Bessel Null method, which is the best way of doing it, without having to know most (if any) of the maths behind it. You really have to be an idiot to cock it up after the initial setup. I've put DJs that aren't happy unless the entire VU meter is lit up permanently through it and it just carries on attenuating the input, doesn't complain, and doesn't over-deviate. Another nice feature is that it appears to have a smoothing algorithm - that is, if you take a recording that is clipped by analogue hardware, Breakaway does a good job at "rounding" the output.

If you set it up properly with a spectrum analyser, (granted, you have to play some pretty horrible test tones whilst on air) you will not over-deviate no matter how hard you drive the mixer relative to the set level. I'm not sure what happens when you start clipping the sound card input and get digital clipping... I'm not sure whether or not the farting goes up the band as I've not had an urge to try it live, but I will when I take some direct measurements as it's relevant.
The workaround is to set the level such that the red lights going off the top of the mixer don't clip the sound card input, as Breakaway's AGC works as a decent, relatively fast expander if set on the appropriate preset. Noise floor isn't really a problem as a result of doing this, as you're going to deliberately at least halve your dynamic range anyway. Alternatively, or additionally as I prefer to do it, a 3:1ish compressor wouldn't hurt before the A to D to just take the edge off of those peaks.

[Albert H]

I've got Breakaway running on an HP desktop machine with a Soundblaster 192kHz-capable card.

I connected it directly to an exciter. My standard exciter was set to 98 MHz - nominally the middle of the band - and was connected through a sampler head into a 50Ω load. My HP2732 analyser was set up to look at the close-in products either side of the carrier. With no modulation input, there were two synthesiser artefacts at -105dBc and 16.625 kHz either side of the nominal carrier frequency. I soldered on a missing loop filter capacitor, and these disappeared into the noise!

Connecting the sound card with no software running introduced some crud around 1.54 MHz either side of the carrier. This was at -94dBc, and appears to be 8× the sample rate!

Running Breakaway, and the 1.54 MHz was joined by other noises at around -90dBc at 2×, 4× and there was more at 8× the sample rate (obviously neglecting the 19kHz stereo pilot!). There were also harmonics of the pilot....

I wondered if the sound card wasn't really up to the job, so I tried an Asus "Xonar" card, and once I found the right drivers, the results were really horrible! The cheaper Soundblaster won out!

I added a homebrewed elliptic lowpass filter based on op-amp gyrators that goes over at ~90kHz and is way down by 130kHz. This was a scaled version of my 15kHz audio filter, and worked well. All the soundcard artefacts disappeared entirely. I can only conclude that the output of the soundcard isn't significantly filtered as ultrasonic products usually don't matter!

Incidentally - my Linux-based audio processor, stereo coder and RDS generator gave the same spurs as the Windoze "Breakaway", so the noises aren't due to software issues.

[NOYB]

Thanks Albert - some food for thought there. I've not noticed that, but I've not gone looking very hard either! Maybe pro transmitters have a tight mask filter that gets rid of that issue?

[NOYB]

Alternatively, or additionally as I prefer to do it, a 3:1ish compressor wouldn't hurt before the A to D to just take the edge off of those peaks.

My instincts scream at me never to chain processors. It's only subjective I know, but when I've heard chained processors they never sound as good as a good single processor set up properly. I think good MB processors like to have full freedom to operate over the whole of the audio. Obviously I'm talking relatively fast limiting here - a slow AGC should be fine and definitely better than overloading at the sound card input!

Nice kit by the way!

[Albert H]

NOYB - I don't mind "chained" processors, and that's often the way I've done it - separate units for each part of the processing.

My processing rack started with a slow-acting AGC, installed to handle the very worst excesses of pre-recorded tape (or CD) programmes. This would bring the under-level recordings up to roughly where they should be. With some processing, and particularly where the American 75µS characteristic was being used, a sibilance limiter was a good idea. This was an additional part of the pre-emphasis circuit, and would really make a huge difference to programme material with lots of higher frequency audio content. It also made female voices more pleasant to listen to - many female voices sound "tizzy" or "hissy" on some airchains!

After the pre-emphasis, I'd use a 15kHz lowpass filter. Years ago, I'd use the pre-tuned Toko passive filters. These gave very good results as long as the source and load impedances were correct (not difficult to achieve with modern op-amps!). These modules were dropped from my gear for two reasons: Stephen Moss bought the last 2000 of them, so they became (largely) unavailable, and I found that I could get significantly better results from an op-amp, gyrator-based elliptic filter. The circuit that I came up with (which I'm happy to put up on this site), was substantially flat to 15kHz with virtually no phase errors (unlike the Toko filters!) and then fell off really steeply, so that there was virtually nothing getting through at 18.5kHz.

The last item was the limiter. It was essential that this never allowed over-modulation. Most commercial designs guarantee this by using diode clippers - if the audio goes over-level, the diodes "top and tail" the signal, adding gross distortion, but preventing adjacent channel interference.

A conventional feedback limiter cannot prevent overshoots and over-modulation, simply because it has a finite "attack" time - the time it takes for the sidechain to rectify the audio to derive a voltage proportional to the level, charge the capacitor(s) in the time constant circuit and for the gain cell to react to the change in the control voltage. This response time problem is handled in two ways - by using diode clippers (with all their attendant distortion) and by designing the time constant circuit to dump as much charge, as quickly as possible, into the timing circuit.

There's a further problem - the feedback limiter will always overshoot: it'll pull the level down too far, and then "realise" its mistake and allow the level to rise somewhat - this is the infamous "pumping" effect (common to all crude processors) that makes listening to most commercial stations so unpleasant. Pumping and overshoots are common to all "real-time" processors, irrespective of their price. Orban, Innovonics, and almost all the other broadcast processor manufacturers claim spectacularly low distortion and noise figures for their gear. These figures may well be true - for steady-state signals! They all turn to crap when you put real audio through them instead of tone generators!

The "real-time" processor manufacturers have used all sorts of circuits in efforts to overcome the limitations of their flawed approach. They use multi-band processing (one model I saw last year used 26 bands to try to overcome the problem), they use "distortion cancellation" (it's better not to distort in the first place!), and they claim all sorts of magic in their crude clippers and filters to try to hide the fact that they're just distorting the signal!

All these expensive broadcast processors sound dreadful - no matter how carefully they're set up. Some people claim that the various Orban products can be set up to "sound good", but I've never really heard it, and they're generally quite under-modulated compared to the heavily-compressed norm.

The only way to avoid the use of the distortion-generating clippers is to introduce a delay line into the audio. This can be done in three ways - each with their own strengths and problems:

The oldest way of generating the delay was with a series of "phase delay" filters. These were a whole load of paralleled filters spaced across the audio bandwidth, and dimensioned to introduce the same delay at all frequencies. These filters can be damped LC tuned circuits or op-amp all-pass filters. The problems with either "all-pass" approach is that it's difficult to dimension the filters to avoid phase errors and variable delays across the audio band. You also have to use expensive, high accuracy components (1% capacitors and resistors) or spend ages measuring and sorting through bags of components!

The next method is the digital delay line. The simplest approach is to use a "Delta Modulator" and a shift register feeding a simple DAC. This works well, and the sample rate can be well above the audio band (800kHz was the rate I chose) so that any artefacts are well away from the audio. You need good filtering on the way in and the way out, but this is a good, cheap approach, and doesn't require many high precision components. You have to be careful to filter out the switching signals, and make sure that they don't get on to the supply rails.

The final approach - and the one I use most - is the analogue "bucket-brigade" device. These were developed for all sorts of devices that needed audio delays, and as long as you're not asking for very long delays and assuming that there's some pre-processing going on (they're poor with low-level signals), the results are low noise and low distortion. I use the Panasonic MN-series delay devices, with a CMOS-based clock (rather than the Panasonic MN3101, which can't really go fast enough). I use simple Bessel Filters on the way in and the way out, and introduce just enough delay to overcome the (fast) attack time of my limiter.

If the delay is 1ms, it's the same as being (roughly) 1 ft in front of a speaker. If you use up to a 2.5ms delay, it's virtually unnoticeable to the DJ, and gives plenty of time for the rectifier and the gain cell to react to the over-level.

The limiter doesn't ever pump, doesn't distort, doesn't allow over deviation and removes the need for diode clippers. My exciters and link transmitters do still include modulation input clippers - as protection, so that plugging in a modulation source won't ever result in an over-modulated "splat" across the band as a blocking capacitor charges or discharges, and also provides protection to the onward parts of the circuit.

I'm just tidying up the circuits of the processors I use to put them up on here. I really don't mind giving away my "trade secrets" if it makes stations sound better and gets their mod levels right. With these circuits, you can sound better than any of the commercial stations, with lower distortion and clean, loud programme. These circuits have been used in successful stations all over the world, and the modules are usually built to order. I'm NOT taking any orders at present!

[NOYB]

Thanks Albert - very informative.

It's a very subjective subject though. I agree many stations sound horrible with that ear-bleeding wall of sound, but I've also heard several that sound lovely and plenty loud enough. When I've been able to discover the make, it's usually an Optimod.

Maybe my ears are tuned to that Optimod sound, and admittedly I have not (knowingly) heard one of your processors (any of them on-air in UK?).

Also I know typical digital processors employ "look-ahead" limiting - which I've always presumed is achieved by using delays like you do?

I get everything you say scientifically, but subjectively I still think digital might be the way to go for people on a limited budget, and/or with limited electronics building skills (like me!). Heck I believe even Radio 3 use a digital processor, so they can't be that bad!