Test equipment needed for compressors

[JohnRoberts]

Lots of activity since I last visited.

My burst on time was much shorter than a second. Attack effects will settle pretty quickly. I needed to make mine completely adjustable so I could better coax out artifacts in ancient circuitry. Modern parts are quite a bit better behaved than what I was working with.

Brad: yes, my mix control bypasses the burst with a dry signal. This has two uses. Yes, the baseline level is useful to gauge release time, but also establishes initial conditions at some level. The gain cell/VCA and side chain ramping to final conditions from 6 dB away will be quite different than from 26 dB away.

My preference for sine wave burst testing for design listening tests is because the simplicity (purity) of sine waves makes artifacts stand out more. Admittedly this is more focussed on functional integrity of dynamics chain and not aesthetic which must be established with music and ears only.

FWIW my hardware burst generator could also accept music as source material so I could hype the dynamics present using normal music, and gauge how well music masks artifacts that couldn't be completely eliminated, or when forced onto a tradeoff between how the artifact will express. Of course since the burst timing was uncorrelated to music the utility was limited with musical sources.

IMO overall time constants should be set entirely be ear, and maybe measured after the fact with tones. The nitty gritty of design benefits from being able to zero in on sonic consequences of different tradeoffs. Judgement is involved in how much the music will mask things audible with simple tones. I don't recall ever encountering an artifact that was more audible with complex music, but it is entirely possible to not check in all the useful frequency ranges, so pretty much like so many audio tests the simpler stimuli increases sensitivity for a narrowed type of behavior. So listening tests with music can help point you toward the proper specialized test conditions to isolate whatever you are suspicious of.


JR

 

[bcarso]

This all reminds me of an event early in my checquered career at the evil empire, when an ambitious new manager/****** type invited all and sundry to a presentation by some digi-phile acquaintance who claimed to know all about signal processing for audio and in particular, compressor/limiters---and why we were all in the stone age for not using wonderful new and obviously superior digital techniques.

No less that Sidney Harman himself came into the meeting, to the shock and awe of all. It would have been interesting to see a display of blood pressure readings as the meeting proceeded :razz:

At one point while the presenter was showing really simple-minded graphs of how his wonderful digital compressor would work, consisting of straight-line segments and no discussion of dynamic response (or even the need for it!), I had an opportunity to comment. Although it was something of a cheap shot, I said "The subject of compressor/limiters is vast". This elicited a supportive sort-of-guffaw from Mark Gander. The presenter was undaunted, and after a while it became pretty obvious that he was at best naive, and at worst an idiot.

Remarkably, the ******/organizer lasted for another year or so.

 

[JohnRoberts]

That reminds me of a presentation we got years ago at my last real job, where a new "digital" power amp company gave us a presentation on the future of power amps (ignoring that we had been making and selling a class D amp for maybe 10 years at this point).

The scariest thing about the meeting was they didn't even grasp the significance of some of the questions we asked them. They went public after that so probably had quite a war chest to burn through. I just went to their website and there is no new press releases activity since losing some investor lawsuit in '06, and numerous broken links on website. Hmmm.

JR

PS.. perhaps ironically one of the best downward expanders (a glorified noise gate) I ever heard was a digital algorithm generated by an unremarkable engineer. When I asked him to explain how he was doing it I could not make sense of his explanation (perhaps a personal problem). The downward ex was added at my request to band-aid a noisy multi-efx unit.

 

[bcarso]

[quote author="JohnRoberts"]
PS.. perhaps ironically one of the best downward expanders (a glorified noise gate) I ever heard was a digital algorithm generated by an unremarkable engineer. [/quote]

I think it was some spinoff of Orban who presented to Harman a few years ago a bit of firmware, maybe some hardware as well, for dynamic processing. It sounded pretty good for the horsepower required, although the examples used may have been carefully tailored. The claimed genesis was the work of a Scandinavian teenager doing quasi-pirate radio broadcasts, who couldn't afford any of Orban's commercial equipment so rolled his own---using purportedly a 486!! He corresponded with Orban and they hired him.

I hope the story is true, as I was quite impressed. It wasn't perfect, but did awfully well considering that it introduced only about a 2 ms delay.

 

[JohnRoberts]

I need to make an uncompressor, to turn down commercials... I leave a business news channel running in the background.. but making it loud enough to understand the soft speaking talking heads makes the commercials unbearable. Some can be quite annoying,, if you're not in the market to eliminate unsightly belly fat.

My favorites are those modern exercise machines that give you bulging muscles effortlessly. :?:

JR

I suspect DSP could make a very smart compressor, and I don't fully agree with some of the basic principles that current dynamics processors follow. They were probably driven by early hardware considerations and just built upon due to lack of imagination combined with the marketplace resisting change, but the world doesn't need yet another comp-limiter, when I have a wide open field to plow.

 

[barclaycon]

JohnRoberts wrote

Back when I was designing companding NR and compressor/limiters, I made my own tone burst generator. Basically a glorified gate circuit, that would gate some other sine wave or signal source as desired.

That would make a great project John.
A tone burst generator is one bit of kit I could do with. I'm sure a lot of others here would find it useful too.
I tried to line up a companding circuit in a stereo simulator unit the other day and it specified a tone busrt generator for lining up the NE572 chips.
I'd also love to see what the attack and release (particularly attack) is on some units I have. EMT 156 for example, where they have an internal trimpot to adjust the attack but no description in the manual of how to adjust to the figure they specify.

Martyn

 

[JohnRoberts]

[quote author="barclaycon"]JohnRoberts wrote

Back when I was designing companding NR and compressor/limiters, I made my own tone burst generator. Basically a glorified gate circuit, that would gate some other sine wave or signal source as desired.

That would make a great project John.
A tone burst generator is one bit of kit I could do with. I'm sure a lot of others here would find it useful too.
I tried to line up a companding circuit in a stereo simulator unit the other day and it specified a tone busrt generator for lining up the NE572 chips.
I'd also love to see what the attack and release (particularly attack) is on some units I have. EMT 156 for example, where they have an internal trimpot to adjust the attack but no description in the manual of how to adjust to the figure they specify.

Martyn[/quote]

While I gave a written description of the circuit a few posts back, perhaps I'll scratch out a schematic. I am always apprehensive about publishing a schematic that I haven't personally tested for errors, but I'm too busy to prove it out, so any new schematic I generate (if I do it) will be posted as is. Maybe next weekend.

JR

 

[barclaycon]

Great.
Look forward to it.

 

[JohnRoberts]

OK, thats a few hours I'll never get back... Here is a rough schematic for a burst gate, that will take a sine wave input and allow you to gate it into a useful tone burst.

This circuit has not been bread boarded and I left a number of values un specified, so you can substitute what parts you have, and tweak others to taste. I typically used a tl074 for the opamp, probably a J-111 or similar for JFET.

The gate on/off time constant circuit is not overly sensitive to pot values and audio taper may give a better adjustment law. Pan pot between gate and straight should be linear and too low impedance will change how it acts while panning but no effort was made to maintain constant power or voltage.

For my convenience I used +/- 15v supply. The cmos FF is connected from ground to -15v to give appropriate gate drive for N ch J-FET. To use with P ch, move FF to +15V and gnd.

I'll be glad to answer general questions, but would rather you fill in your own values for time constants. While I probably used a big mylar and Meg+ value resistors, you could use a low leakage electro for gate on/off circuit as ramp linearity is not an issue.

Enjoy

JR

 

[barclaycon]

OK, thats a few hours I'll never get back...

Think of it as an investment John !
It 's much appreciated, believe me.
Just as a matter of interest, do you think this circuit could be applied to the following alignment procedure for a NE572 compander:

The compressor attack time is speeded by a charge pump, which reduces transient distortion. To align the circuit set the tone burst generator to input a signal to the unit of 1kHz at +4dbM and set the gate duration for 32 cycles open and 32 cycles closed. Set the charge pump preset value for minimum overshoot or undershoot of the first cycle.

Look forward to your comments.

Martyn

 

[JohnRoberts]

Investment? I'll wait for the Karma payback.... (waiting).

Yes, this is very similar to what I used to do a great deal of design with 570, 572, and real VCA based dynamics processing. My old unit just added in the direct signal in parallel instead of panning to get a burst depth. Another approach could be a variable R in series with the shunt JFET to ground.

A 32+32ms, 50% duty cycle burst is not my first choice to tweak a 572, but they are probably assuming a more primitive tone burst capability and using the predictable amount of control voltage release during that 32mS off time to establish the initial conditions for their burst adjustment. This is roughly equivalent to a say 5 dB burst step above a fixed nominal level with that burst occurring at a much slower, say 1 sec rep rate.

I did a certain amount of research in trying to correct CV feed-through in 572 and could never get a perfect null over entire operating range, so always edge limited CV signals to prevent clicking. Pre/de-emphasis also mitigates fast gain change artifacts.

JR

PS: for that other poster looking for things to do with a VCA, he could make a burst gate. :wink:

 

[Kid Squid]

Hello Chaps,

Wow, got a lot of stuff to digest here :shock: , haven't been around for a while, and I'm amazed at the response to this post.
Thanks guys, this information is highly respected by myself, my hat goes off to you :guinness:

bear with me.....

found a schematic of a tone burst generator http://www.akh.se/myown/7304.htm

haven't built it yet, and in response to the posts, thinking of using my DAW to do the testing.
have you got anywhere yet Tom ?
So , Thanks again chaps, off to print the pages out - have a read on nights tonight.

Iechyd Dda,
( overwhelmed at the response, I am)

 

[TomWaterman]

Damn sorry Squid - I didn't see your previous post last time around. Yes I got somewhere. Coming back to this thread to post an intersting result...at least for me it was.

The test files I used are a bit ****** (due to poor software signal gen) and are only steady-state testing at various frequencies...but its pretty good to show attack and release behaviour and very good at viewing distortion of low frequency waveforms caused by fast attack and release times.

Measured some plugins and some Drawmer, FMR and other 'nameless' hardware...for now.

The FMR RNC measures quite well in that regard (LF distortion)....supernice mode really smooths out distortion at 20Hz by slowing the attack and release and limiting the amount of G.R. It is also very close to the panel labelling for attack times etc - due to the DSP sidechain. Some of the analogue devices were very variable and often matching between two channels was awful.

The test files I used are here - DivShare link - not long enough for 2+ second releases but easy enough to make a new burst tone with a software sinewave generator. There's some program material in there too, orchestra and snare...

I used 20Hz, 100Hz, 1kHz and 10kHz...but settled on the 20Hz and 1kHz being the most useful. 10, 20 and 30dB ranges...I found 30dB best for attack times, set for heavy G.R and 20dB best for viewing release times...but bear in mind all the interaction between controls and amounts of G.R etc.

The following plot is from a hardware compressor which does stuff I had never 'seen' before during the attack/release high amplitude section, where most normal compressors (at least the ones I measured) remain in G.R until the input signal drops below the threshold again.

Any guesses as to whats up with this one (the last two shots)? It's a feedback, fixed threshold F.E.T compressor...attack and release were set very quick (uS range for attack..) for an indication the 1176LN didn't do this. Some sort of attack/release interaction? The actual manufacturer has no idea!

-Tom

 

[JohnRoberts]

While it's difficult to say exactly what's going on, it appears to be an artifact of too fast attack and a nonlinear gain law.

At the initial burst it says drop the gain a huge amount, but since it's wired as a feedback limiter it realizes it overshot and ramps up to a new equilibrium. One way to think about this is it's so fast, it's responding to what the signal was, rather than what it is. Half wave detection will also reduce the ability of the control loop to reflect reality quickly.

Using a FET as gain element will never deliver a well behaved gain law. The feedback topology is used so the gain law is not very important as it limits around the threshold. You just need to slow down the attack time to something that doesn't overshoot so obviously.

While I prefer stepped or dual level bursts as a better mimic of music for dialing in dynamics, consider maybe some loops of snare hits, or bass slaps to dial it in for real world signals.

It will alter the sound character, but hopefully not trash it completely once you slow it down some.

JR

 

[TomWaterman]

OK thanks John - interesting right enough.

It sounds pretty good this device, even set to the fastest attack, although it does sound pretty 'pancaked'....attack and release seem very interactive and balancing them is the key to get it to perform musically. I was pretty certain it was something to do with the speed of the attack and feedback topology.

I have schematics here but have had no real time to study them...I'll check and see if it half-wave rectifies the S.C

I have used drums, orchestra, snare and the tone bursts as test sources.

Whats interesting is the 'hold' in the attack. As you back it off it doesn't really do anything for a good few mS and then it reduces level - you can see in the bottom pics as the attack is adjusted slightly slower.

All of the other compressors I measured start to kick in as soon as the high amplitude section hits and the attack time just elongates the reduction to stable compression. This one stops the thing from reacting and then it squeezes hard.

Thanks again.
T

 

[JohnRoberts]

For evaluating attack time consider using HF sines, since lower frequency sine waves may have periods longer than attack times.

The bottom line for dynamics processing is what does it sound like, but IMO FW rectification will be more consistent.

JR

 

[TomWaterman]

The above is 1kHz.

I ran some 10kHz as well.

20Hz is nice to view the distortion...some comps get ugly!

The above compressor was designed by ear and the manufacturer has no idea how it behaves with regards to these tests...BUT it does sound pretty cool for a sub £500 device.

Nice on drums.
Cheers Tom

 

[JohnRoberts]

The LF sine wave can reveal distortion caused by fast release rates if there is no hold time. A popular finesse for this distortion mechanism is to add a couple tens of mSec hold time before you start releasing. In general this is not problematic since release times typically run much longer than that.

In practice complex waveform envelopes rarely look like pure sine waves but for say compressing a bass track or synth there could be audible distortion, with very fast release times and simple control paths.

There are sundry tricks for managing ripple in control path which is a source of distortion.

JR