SNR figures for VCA based compressors.

[rodv92]

Hi.
I am designing a compressor based on THAT2180A VCA from scratch.
No previous experience in compressor design.
I used the approach "try to fully figure it yourself and then confront to real life designs"
I used Giannoulis paper
"A Design of a Digital, Parameter-automated, Dynamic Range Compressor" to get the building blocks of forward based gain computer for the sidechain.
It is for a digital implementation so I did not get influenced by any analog schematics.

So far, so good using Ltspice. I got some deviation from the behaviour of a theoretical compressor, probably due to the op-amp log amplifier not behaving as the ln() function, even after renormalization.
Mostly happens with low thresholds and low amplitude signals, which is expected. So, yes calibration was already hard in simulation.

Now I am into the thermal noise simulation. Using ltspice noise analysis, I get 132dB SNR using a sine wave at 4dbU (1.736Vmax) at the input and measuring noise at the end of the sidechain, just before the THAT2180 VCA. and 201dB SNR at the audio path output.
I expect that in reality intrinsic SNR is higher due to parasitics, loop area, routing techniques, supply noise, grounding and shielding, crosstalk, etc...
A friend of mine gave me the target of 110dB SNR, and indeed this figure is frequently found in audio pro level products.
Could you enlighten me and give me some figures for state of the art VCA designs (modern as well as vintage) ?
From your experience, shoud I be overly concerned by it or will it be another factor of concern when going from design to prototype ?
It may be hard to tell if I don't publish the design for now, but let's say its quite heavy in the number of op-amp stages in the sidechain : 16
and a parallel branch of 8 stages for hard/soft knee management.

Any advice appreciated.

 

[Bo Deadly]

LTSpice is great but there are no models that will accurately reproduce the behavior of a THAT2181. Blackmer VCAs have traditionally been quite noisy. It's only because of the special dielectric isolation fab used by THAT that the THAT2181 is able to achieve the noticeably lower level of noise and THD that everyone else. So LTSpice is just not going to help you with noise and THD sims. I also sincerely doubt that an LTSpice simulation will actually get the dynamics right. There's just too much going on in that circuit to model it with the conventional parts available to sim.

The best path IMO is to go straight to PCB. You can't even really breadboard these things. One of the most influential parts of the circuit is the noise and distortion at the control ports. Any noise / thd will directly feed-through. You really need to be careful about the layout with respect to the control ports. You want very short traces and very low impedance. Look very carefully at the datasheet about layout and at design notes for any clues about how to best control the port(s). That will be where one implementation does better than another.

 

[rodv92]

Thank you for your advice.
Indeed the THAT datasheet is quite well documented.