Pellucid Compressor: an ultra low noise compressor for guitar and bass

[jonny.reckless]

Introducing the Pellucid compressor - a new original design stompbox compressor for guitar and bass. Unlike the Reckless diode compressor, which was a feed-forward design, this is a more traditional feedback compressor design with a two transistor peak detecting sidechain that is derived from the DOD 280, as used in just about every other guitar compressor out there including the Dynacomp, the Ross compressor, the DOD 280 optical compressor, the Diamond optical compressor, the Mooer yellow comp and lots of boutique clones.

The Pellucid uses an LM13700 in the feedback loop of a NE5532 low noise op amp for its VCA. This keeps background noise down to an ultra low level. This is the same basic technique for low noise that is used in the Engineer's Thumb compressor. I've paralleled up both halves of the LM13700 for a further noise reduction, and drive the OTA with a balanced differential signal to minimize control signal breakthrough.

The design goals were simplicity, transparency and low noise. It has a small component count and is made of standard parts all cheaply available from Tayda electronics. It fits into a 1590B enclosure. There are just 2 controls: sustain and volume. Attack and release time are fixed, and both pretty fast, suitable for a wide range of electric guitar styles. The amount of compression available goes from very little (essentially a clean boost) to up to 40dB of gain reduction, which is a massive squish even with single coil pickups. It remains quiet as a mouse even at extreme compression settings due to the architecture and use of a low noise op amp. You could substitute an LM833 or LM4562 in place of the NE5532 if you wish, although the improvement would probably be negligible. I'd avoid the TL072, RC4558 etc as they are generally too noisy for this application.

Hear it here:

I have designed a PCB for this compressor. It's a 2 layer board, carefully laid out and optimized for low noise operation. It is 50mm x 55m in size. If you want one, I have a few spares and can post you one for $5 plus postage.

Here is a link to my Google drive folder, which contains the schematic, bill of materials, several photographs of the unit, and the PCB Gerber and NC drill files should you wish to etch a PCB yourself:
https://drive.google.com/drive/folders/1p9QCJDEhjd2qE1EpuAr-Ndzg3K435dOq?usp=sharing

I'm very happy with this compressor design. It's simple, cheap, easy to build and sounds great.

Jonny

[Ben N]

Jonny, you're killing me! I just finished your diode compressor (v2), and now this. It sounds incredible, although maybe the difference is not that great that I have to make it. The diode comp is also plenty quiet for my ears, and is everything I expected it to be as a clean front end. How would you compare the two sound-wise?

[Elijah-Baley]

This compressor seems really nice. I would like to try it. I'd need a veroboard layout, because I build using that stuff, maybe I'll try to make my own layout.

I'm not a fan of compressor pedals, one of my first pedal I built was the Orange Squeeze, just because it was a simple one, but for some reason it was enough volume, I need to get mexed. After some years I tried to set the internal trimmer, but I can get enough volume without to cause distortion. (I added a tone pot in the end of the circuit, but I'm not sure this casued the volume drop.) Maybe I'll built it again, just to not lose the box drilled and labeled. And, I liked that squish sound, it was kind of funny to play with it.

Now, after some months, I finally boxed up the Engineer's Thumb, but, though I built two boards, I had some issue about a louder attack of the first note at certain setting of the Release and Threshold control, that causes even a loud click noise when I act on the footswitch. I partially fix this, but is not perfect, especially at high compression settings.

So, I would like to try another compressor.

[jonny.reckless]

How would you compare the two sound-wise?

The diode compressor is more transparent due to its feed-forward side-chain. It's more like a studio compressor in terms of its dynamics response. The pellucid is more like a dynacomp in terms of its feel, albeit much lower noise and with faster release. It has that classic quack, and you really feel the compression when playing, since it is acting like a limiter rather than a traditional ratio compressor. Both have their uses. The engineer's thumb, diode compressor and this each have their own character, and they complement one another quite nicely IMO. You can probably assemble a pellucid in a couple of hours; it's a simple circuit. You have to be careful building this on veroboard, because the high loop gain around the 5532 / 13700 means it tends to oscillate unless it is well grounded and parasitics are kept to a minimum. The PCB took me a couple of iterations to get working really cleanly.

[Eb7+9]

The diode compressor is more transparent due to its feed-forward side-chain. It's more like a studio compressor in terms of its dynamics response. The pellucid is more like a dynacomp in terms of its feel, albeit much lower noise and with faster release. It has that classic quack, and you really feel the compression when playing, since it is acting like a limiter rather than a traditional ratio compressor.

A servo AGC loop should give you more stable behaviour  ...

In looking at your schematic
You’re driving two low-z current inputs (OTA biases) thru 10k scaling resistors
from a high-z cap and resistor combo

Yielding a very short release time in the process

I suspect your circuit is operating on the verge of acceptable
in that respect

If wanted you could buffer that control line with a rail-rail opamp,
where voltage turns to current,
and have a longer (and variable if it matters to anyone) release time

your impression of feed-fwd and feed-back might change once you do that

—-

again a very original and interesting design effort
the only limiter I’ve seen add current back into an opamp (remotely) this way
is the SWR circuit

[jonny.reckless]

I suspect your circuit is operating on the verge of acceptable in that respect.

I think everything is OK? I've tested it on the bench with input signals from 10mVpp to 3Vpp, from 80Hz to 8kHz, with continuous tones, bursts, pink noise, and step amplitude changes, and it seems to be working acceptably. Attack time is level (and frequency) dependent but nominally measured at 3ms, release time around 200ms.

As configured, each OTA requires approximately 400uA into its bias pin to hard limit the amplitude at maximum compression with a maximum signal level of 3Vpp. This causes the collectors of TR1 and TR2 to rise to about 5V at maximum gain reduction, due to the 10k current sharing resistors, well short of saturation. Release time is set by C1, with 47uF giving a release time of around 200ms measured on the bench. I quite like fast attack and release for guitar compression, it makes it feel more 'alive' when you are playing, especially with single coils. Most dynacomp style compressors have way too slow release for my taste. You might want to increase C1 to 100uF for a bass compressor though.

[merlinb]

I've tested it on the bench with input signals from 10mVpp to 3Vpp, from 80Hz to 8kHz, with continuous tones, bursts, pink noise, and step amplitude changes, and it seems to be working acceptably.

I'd be interested to see the gain curves?

[jonny.reckless]

I'd be interested to see the gain curves?

Do you mean input dBV vs output dBV steady state? Or dynamic response to a step change in amplitude? I'll have to figure out how to do a screen capture on my oscilloscope for the latter

I'll see if I can do that at the weekend.

[merlinb]

Do you mean input dBV vs output dBV steady state?

Yes steady state. Not necessarily in dBV, ordinary V will do.

[bool]

Ha. Fight of the giants has started!

Heh heh. Are we witnessing the beginning of a "pellucid thumb" compressor?

[Mark Hammer]

While I doff my cap and bow deeply and humbly at a thoughtful and generous design/designer, and appreciate every diligent attempt to reduce noise WITHIN a circuit, too many observers and novices misattribute the "noisiness" of a compressor to the compressor itself.  In reality, the compressor is simply doing what you asked it to do: amplifying low-level signals and trimming back high-level ones.  Realistically, one of the best things one can do to minimize noise in a compressor is simply feed it the lowest-noise signal you can.  Those with any experience will likely respond to this with "Well DUH!".  But every day there are thousands more newcomers who don't understand how compression works, and need to be reminded/informed of such things.

[bluebunny]

But every day there are thousands more newcomers who don't understand how compression works, and need to be reminded/informed of such things.

My tame guitarist still thinks that "sustain" is a thing (other than the physics of his guitar).  I've explained to him how a compressor actually works, but no, the world is still flat... 

[samhay]

Seeing as everyone else is here, I'll join in to.

Nice idea. A few thoughts:
1. low noise and a NE5532 biased via a 1M resistor are not entirely consistent.
2. you don't need separate resistors (R9,10) for the LM13700 bias pins. Buried in the app notes is mention that these are balanced and see e.g. Fig 32 in the application note (below).
3. you're not using the diode bias. I know that Merlin isn't convinced these are worth the bother, but I'm not sure I entirely agree.
4. R11-13. How well balanced are the amp inputs? The use of trimmers in most of the app notes suggest you may not be getting very good balance/cancellation with this setup and/or the ORIGINAL engineers were overzealous (again).


http://www.ti.com/lit/ds/symlink/lm13700.pdf

Edit - +1 on seeing the static gain curves please.

[jonny.reckless]

1. low noise and a NE5532 biased via a 1M resistor are not entirely consistent.

Thanks for your comments. I'm thrilled people are paying such detailed attention!

From a noise perspective, the 1M input resistors are in parallel with the guitar pickups at audio frequencies, typically about 5k - 15k,  which reduces the noise considerably. I agree 1M alone would be noisy as hell, as you could hear if you ran the device flat out with the input open circuit. The compressor is really very quiet; quieter than any of my other compressors (I've got about a dozen) with the exception of the Empress which is about the same. Noise in compressors is tricky to quantify because it depends on gain reduction, output volume, input impedance, bandwidth of measurement etc but it really is "low noise"  I can run this into my amp with the gain cranked and I don't hear any obvious increase in hiss when I kick the compressor on, unlike the Dynacomp which sounds like a steam kettle under the same circumstances. Bear in mind that with little or no signal, the LM13700 is off, and contributing no noise to the circuit, so it's just the noise gain of the opamp based on the sustain setting. NE5532 is a quiet opamp when driven from low impedance sources. The voltage noise and current noise equal out about 22k ohms input impedance. With my strat pickups the voltage noise dominates. If you switch the opamp for a TL072 you can clearly hear the hiss level increase compared with the NE5532. I really like the NE5532 and NE5534 opamps for audio - they're cheap, quiet, and can drive quite low impedance loads with low distortion. I only use TL072s when super low input bias current is absolutely necessary. LM4562s are nice but not as robust as NE5532 - I've blown a lot of them up while prototyping, which almost never happens with a 5532.

It's interesting you don't need to current sharing resistors according to the datasheet, because the voltage on those pins differs by about 10-15mV on the bench, enough that one of them would hog most of the current if I hard wired them in parallel (~60mV per decade re Ebers Moll). I guess they even out as the current increases? Generally I use current sharing resistors anytime I have bipolar transistors in parallel, as in this case (the bias pin looks like 2 diodes to ground). I'd certainly want current sharing resistors if I were going to build that state variable filter from figure 32, you need close matching of the transconductance between the two halves to get the poles coincident. For the cost of one extra resistor (a penny or so) it's worth it to me anyway 

I've never really seen much benefit in the linearizing diodes. It makes biasing harder, and the few times I've tried it, I couldn't get the quoted 10dB in noise reduction at the same distortion level.

You're absolutely right re R11-R13, trimmer balance here would be better, with 1% resistors I am limited to 40dB CMRR best case. I just decided to run a balanced signal into the OTA because I already had an inverted signal to drive the transistor peak detector, and it seemed sensible to use it. Whilst not perfectly balanced, the common mode input signal will be much smaller than if I didn't drive it this way, so you don't really lose anything. I find LM13700 don't behave that well at 9V or below (the output impedance is asymmetrical +ve and -ve going), they really like 15V or more, so anything you can do to help them out is probably a good idea. For a guitar effect the asymmetry generates even order harmonics which is not necessarily a bad thing.

[jonny.reckless]

Ha. Fight of the giants has started!

No fight here. I have the utmost respect for Merlin and others on this forum (his valve amp book is great BTW). The Engineer's Thumb compressor is a great design; I have one on my pedalboard right now. The pellucid and the diode compressor fill a niche which I thought was missing.

I'm using the "OTA in the feedback loop" approach from the ET, and the "2 transistor peak detector" sidechain approach from the DOD280, Dynacomp etc. I just married them up to make an original circuit from bits of other designs. But it really is very quiet I promise 

[jonny.reckless]

Here is a link to the static gain curves of the Pellucid compressor with the sustain set to minimum, 12 o'clock and maximum:

https://docs.google.com/spreadsheets/d/1g14rXTT5gZJL6CVnEW8s0eTMZVhcRcZsx16u4_0d22k/edit?usp=sharing

As you can see, it's basically acting like a limiter (high ratio) with variable threshold, rather like the Dynacomp, Ross, Boss compressors. The two transistor peak detecting sidechain clamps the output pretty hard. I added a 10k resistor in series with each transistor base, which softens the knee and allows a little bit of slope in the limiting region.

[merlinb]

Just noticed something on the schem: R16/17 aren't really 1k are they? (Consuming 4.5mA!)
Also, I'm thinking you could use one of the on-board Darlington transistors to buffer your side chain from R9/10, which would give you access to longer release times.

[bool]

Ha. Fight of the giants has started!

No fight here.

BS. ...

We want fight!
We want fight!
We want fight!

heh heh


OT. wrt darlington; yes of course; but if you aim for the uber-gains etc it may be better (layoutwise) to use a separate discrete darl. to split the control path from the signal path powerwise. (theoretically maybe perhaps)
And a darl. will drop how much from the control electrode (b-to-e)? 1V or so... a small mosfet could be even better if it works but will drop g-to-s 2-or so volts. Can you live with that?

[samhay]

Thanks for your comments. I'm thrilled people are paying such detailed attention!

You're welcome.

>From a noise perspective, the 1M input resistors are in parallel with the guitar pickups at audio frequencies,...

Indeed, that's a valid point if you put the compressor at the start of your signal chain. Not quite so good when it goes after a Fuzz Face. It's a compromise, and I've certainly designed things with an NE5532 in a very similar input stage.

The other issue though is how much DC offset you end up with. Via the inverting stage, you could have quite a difference in DC levels going into R11 and R13.

>The compressor is really very quiet; quieter than any of my other compressors (I've got about a dozen) with the exception of the Empress which is about the same....

Perhaps the fairest comparision would be against Merlin's ET, which I would guess is pretty similar.

>It's interesting you don't need to current sharing resistors according to the datasheet, because the voltage on those pins differs by about 10-15mV on the bench,...

Thanks for the real world information. Much more useful than most data sheets.

> I've never really seen much benefit in the linearizing diodes. It makes biasing harder, and the few times I've tried it, I couldn't get the quoted 10dB in noise reduction at the same distortion level.

Fair enough.

>You're absolutely right re R11-R13,...


And also fair enough. Interesting observation RE 9V supply for LM13700. I've noticed that too, but wasn't sure how representative it was.

[jonny.reckless]

Personally I quite like the short release time.  It's a gripe of mine that most compressors have too slow a release time for my taste. I think 100ms to 200ms is great for guitar, maybe twice that for bass or vocals? The compromise is getting fast release without too much amplitude modulation at low frequencies. I experimented with using a second order filter in the sidechain but it didn't really sound any better to my ears. But it would be cool to see how a longer release worked with the Darlington buffers. It would however mean that the collectors of TR1 and TR2 would now have to rise to 4 diode drops above ground before any gain reduction occurred which would slow down the attack time somewhat.

Re noise performance: the ET uses a TL07x and one half of the LM13700 in its VCA. The Pellucid uses NE5532 and both halves of the LM13700 paralleled. Also the base resistors feeding the OTA are 10x smaller in the Pellucid (this actually makes an appreciable difference to noise since the signals there are so small). Overall this contributes to making the Pellucid about 11dB quieter than the stock ET at no input with a 5k source impedance (typical strat pickup) and compression on maximum. I was being honest when I said it's the quietest guitar compressor I have (except for the Empress which I believe uses a THAT corp Blackmer VCA). The engineer's thumb redux mods I did, take the ET to the same noise performance (NE5534 with external compensation, both halves of the LM13700, and 1k / 22 ohm base attenuator resistors) and are totally worth doing IMHO since they add little or zero cost and complexity but significantly improve low level noise performance.