DC Transfer function problem...

[Magnus Danielson]

> No your not being stupid, good compressor designs have been some thing that many pro manufacturers have a problem with ( no I won't say who but some 
> "pro" compressor/limiters are not fit to grace a $25 cassette machine).
> 
> The control side-chain can either be feed forward or feed back (dbx tend to use feed forward most others use feed back).  The feed forward loop has some 
> advantages but requires more thought to get it working properly (it does have lots of scope for wave distortion though !).

The feed backward solution has a tendensy to oscillate the amplitude, getting
a "pumping" effect once popular. The feed forward does not have the same effect
(unless intentionally built into the control circuitry... which is fully 
possible).

> What you basically need is a "modulus" device i.e. a full wave or half wave rectifier (some companies just use diodes)/comparator/rms detector as a front 
> end.  The choice depends on what you want to do eg peak/average/rms control.  This is followed by a dead zone circuit (rectifier with offset) to set the 
> threshold (thecomparator frontend includes this function). Add an AD time constant and finally a variable gain stage to set the ratio.

It is noteworthy that some companies use a halfwave rectifier on their 
compressor/limiters. This has the flaw that if the initial (strong) pulse have
negative polarity it will not be detected and the limiter is by design broken.
I have seen this myself and I rebuilt it into a full-wave rectifier directly
(which was the original design BTW... but someone did change the circuit at the
factory.... he later got fired I heard).

When designing a detection circuitry it is allways worth considering how to 
deal
with steady signals, positive peak bursty signals and negative peak bursty 
signal.

Personally I would like to have separate peak and RMS circuitry.

> You can get loads of information from the Analog Devices texts:- Non linear circuits and the Audio Video reference manual (application notes section).
> There are also some interesting patents out there, try searches for Aphex, dbx and Orban.
> 
> Perhaps the trickiest bit is the time constant section as a simple AD network can have a number of operational problems.  The worst or best? is that fast
> time constants will cause distortion of the waveforms  (if the time constants are say 1mS and the input signal is 100 Hz the wave will be distorted).  There are 
> many methods of over coming this From multi-time constants to multiband systems.

Half-/Full-wave rectifier curcuit makes a difference here. The fullwave 
rectifier wins.

Magnus