The Ultimate Homebrew Limiter!
Posted: Wed Aug 22, 2018 1:21 am
Most commercial audio processors for FM (and AM) broadcast use include some kind of clipper. If you're Robert Orban, you split the audio into lots of small bands of frequencies, using insanely over-engineered filters, clip the hell out of each channel, filter again, and then mix the resulting mess back together again as "loud audio".
In the world of electric guitar, most distortion effects do much the same thing - clip the hell out of the signal.....
Call me peculiar, but I don't want my audio to go through a Fuzz Box, no matter how insanely over-engineered the audio processor is. Distortion is distortion, and once it's added to the path, you can't really remove it.
Conventional limiters use FETs, opto-couplers, transistors, transconductance amplifiers, diodes even variable mu pentodes, and all manner of other means of attenuating audio controlled by a DC voltage. This DC voltage is derived from the instantaneous level of the audio passing through the limiter, usually by means of some kind of rectifier in a sub-circuit called the "Sidechain". There's usually some kind of time constant circuit involved to prevent the DC Control Voltage from pumping up and down, slowing the recovery after over-level input, to make the resultant changes in level less obvious to the ear. Here's where the problems start. The response of the DC Control Voltage will have a finite "attack time", so sudden changes in level cause the limiter to allow peaks to sneak through. Attempting to make the Sidechain Attack time constant as fast as possible also introduces the problem of "overshoot", where the gain is reduced too much, leading to "holes" in the audio after peaks. The peaks will cause over-modulation at the transmitter, giving rise to nasty-sounding audio at the receiver and the risk of interference to adjacent stations. Commercial equipment generally includes a clipper to clobber these fast peaks - leading to the Fuzz Box problem.....
Back in my youth, when I was employed (and educated) by a Big Broadcasting Concern, we had "delay-line" limiters that were used to prevent over-modulation of Medium Wave transmitters. They were also pressed into service for levelling telephone calls in News programmes. The delay line (a board full of inductors and capacitors) was a set of "all-pass" filters set at staggered frequencies in an effort to get a reasonably flat frequency response. The delay was measured in the µs, but was enough - the Sidechain looked at the incoming audio in the usual way, but the audio was delayed before it got to the gain control element, so the "Attack Time" of the limiter was effectively zero! The limiter was effectively "seeing the peaks coming" and all the nastiness vanished. There were stereo, 15kHz-wide versions of this built too, but they proved to be horribly expensive, and messed up the stereo imaging because of the variable phase shifts in the delay line....
There were some digital delay lines built - using analogue to digital converters at the input, a shift register (or memory IC) to delay the stream of digits, then D to A converters to recover the audio. Whilst these worked extremely well and gave great results not only for broadcasting, but for disc cutting too, they were incredibly complicated and dreadfully expensive.
Now I'm an old-fashioned kind of chap, and doing the digital conversion, limiting processing and audio recovery in a computer seems a bit like cheating! It can be easily done, and there's even a plug-in for "Audacity" that gives delay-line based limiting.
I set myself the task of building a better mousetrap. In the mid-80s, I worked at Panasonic in Japan. They had a series of "bucket-brigade" delay-line chips for audio. These worked by taking an instantaneous sample of the audio as a voltage across a capacitor, then on each clock pulse passing this voltage to the next capacitor in line, until it gets to the end where a simple lowpass filter recovers clean - but delayed - audio. These ICs were used for time-based guitar (and vocal) effects, such as reverberation, echo, flanging, chorus and phasing.
Most people dismissed the idea of using these ICs for broadcast limiting since they were seen as a gimmick, but Garrard made a "scratch remover" that would "de-click" scratched records. It worked by "seeing the peaks coming" (a large, rapid, common-mode spike is going to be caused by a scratch), and inserting a fraction of second of silence in place of the click. It worked really well, and I've used this method to recover some treasured old records that would otherwise be unplayable.....
The delay only needs to be µs.... The MN3007 is still widely available. It's a bit longer than I'd ideally like (has too many "bucket" stages), but it's the one that we can easily get hold of at a reasonable price. The specification says that the upper clock frequency is ~250kHz giving a delay of ~2ms. In reality, if you don't use the original Panasonic MN3101 clock chip, and buffer the clock lines, you can get it to go at around 1MHz, giving us around 500µs. That's just about perfect.
My prototype used a crystal oscillator module at 4MHz, driving both halves of a dual bistable (4013) to give 1MHz, then a 4049 with the gates paralleled in two lots of three to drive the MN3007 with the bi-phase clock that it requires.
The Sidechain is simple, using op-amps as "ideal diodes" to do the rectification. The attack time of the Sidechain is very slightly shorter than the delay, but this compensates for the time taken for the gain control circuit to respond. I used an LM13700 in the feedback loop of a good quality op-amp. To reduce the op-amp gain, I increase the transconductance amplifier gain, increasing the negative feedback and thereby reducing the output level. The prototype sounds superb - or rather it doesn't. You just can't hear it working! There's no "pumping" or "breathing" and the delay is imperceptible to the ear - it's just as if you're sitting about 20cm further from the loudspeaker!
When I've got all the bugs out of the design, I'll put the circuit up on here. I'm going to reduce the clock frequency to the delay-line chips a bit (1MHz really is thrashing them), and I'll use a cheap standard crystal. There won't be any need to distort your audio to control your modulation level any more!
In the world of electric guitar, most distortion effects do much the same thing - clip the hell out of the signal.....
Call me peculiar, but I don't want my audio to go through a Fuzz Box, no matter how insanely over-engineered the audio processor is. Distortion is distortion, and once it's added to the path, you can't really remove it.
Conventional limiters use FETs, opto-couplers, transistors, transconductance amplifiers, diodes even variable mu pentodes, and all manner of other means of attenuating audio controlled by a DC voltage. This DC voltage is derived from the instantaneous level of the audio passing through the limiter, usually by means of some kind of rectifier in a sub-circuit called the "Sidechain". There's usually some kind of time constant circuit involved to prevent the DC Control Voltage from pumping up and down, slowing the recovery after over-level input, to make the resultant changes in level less obvious to the ear. Here's where the problems start. The response of the DC Control Voltage will have a finite "attack time", so sudden changes in level cause the limiter to allow peaks to sneak through. Attempting to make the Sidechain Attack time constant as fast as possible also introduces the problem of "overshoot", where the gain is reduced too much, leading to "holes" in the audio after peaks. The peaks will cause over-modulation at the transmitter, giving rise to nasty-sounding audio at the receiver and the risk of interference to adjacent stations. Commercial equipment generally includes a clipper to clobber these fast peaks - leading to the Fuzz Box problem.....
Back in my youth, when I was employed (and educated) by a Big Broadcasting Concern, we had "delay-line" limiters that were used to prevent over-modulation of Medium Wave transmitters. They were also pressed into service for levelling telephone calls in News programmes. The delay line (a board full of inductors and capacitors) was a set of "all-pass" filters set at staggered frequencies in an effort to get a reasonably flat frequency response. The delay was measured in the µs, but was enough - the Sidechain looked at the incoming audio in the usual way, but the audio was delayed before it got to the gain control element, so the "Attack Time" of the limiter was effectively zero! The limiter was effectively "seeing the peaks coming" and all the nastiness vanished. There were stereo, 15kHz-wide versions of this built too, but they proved to be horribly expensive, and messed up the stereo imaging because of the variable phase shifts in the delay line....
There were some digital delay lines built - using analogue to digital converters at the input, a shift register (or memory IC) to delay the stream of digits, then D to A converters to recover the audio. Whilst these worked extremely well and gave great results not only for broadcasting, but for disc cutting too, they were incredibly complicated and dreadfully expensive.
Now I'm an old-fashioned kind of chap, and doing the digital conversion, limiting processing and audio recovery in a computer seems a bit like cheating! It can be easily done, and there's even a plug-in for "Audacity" that gives delay-line based limiting.
I set myself the task of building a better mousetrap. In the mid-80s, I worked at Panasonic in Japan. They had a series of "bucket-brigade" delay-line chips for audio. These worked by taking an instantaneous sample of the audio as a voltage across a capacitor, then on each clock pulse passing this voltage to the next capacitor in line, until it gets to the end where a simple lowpass filter recovers clean - but delayed - audio. These ICs were used for time-based guitar (and vocal) effects, such as reverberation, echo, flanging, chorus and phasing.
Most people dismissed the idea of using these ICs for broadcast limiting since they were seen as a gimmick, but Garrard made a "scratch remover" that would "de-click" scratched records. It worked by "seeing the peaks coming" (a large, rapid, common-mode spike is going to be caused by a scratch), and inserting a fraction of second of silence in place of the click. It worked really well, and I've used this method to recover some treasured old records that would otherwise be unplayable.....
The delay only needs to be µs.... The MN3007 is still widely available. It's a bit longer than I'd ideally like (has too many "bucket" stages), but it's the one that we can easily get hold of at a reasonable price. The specification says that the upper clock frequency is ~250kHz giving a delay of ~2ms. In reality, if you don't use the original Panasonic MN3101 clock chip, and buffer the clock lines, you can get it to go at around 1MHz, giving us around 500µs. That's just about perfect.
My prototype used a crystal oscillator module at 4MHz, driving both halves of a dual bistable (4013) to give 1MHz, then a 4049 with the gates paralleled in two lots of three to drive the MN3007 with the bi-phase clock that it requires.
The Sidechain is simple, using op-amps as "ideal diodes" to do the rectification. The attack time of the Sidechain is very slightly shorter than the delay, but this compensates for the time taken for the gain control circuit to respond. I used an LM13700 in the feedback loop of a good quality op-amp. To reduce the op-amp gain, I increase the transconductance amplifier gain, increasing the negative feedback and thereby reducing the output level. The prototype sounds superb - or rather it doesn't. You just can't hear it working! There's no "pumping" or "breathing" and the delay is imperceptible to the ear - it's just as if you're sitting about 20cm further from the loudspeaker!
When I've got all the bugs out of the design, I'll put the circuit up on here. I'm going to reduce the clock frequency to the delay-line chips a bit (1MHz really is thrashing them), and I'll use a cheap standard crystal. There won't be any need to distort your audio to control your modulation level any more!