Im not sure if its ok to ask this here but here goes
Ive been spending time comparing the response of software and hardware compressor emulations. It seems like the best software can get very close in some cases and each improvement continues to impress me. Today Ive been comparing the dbx 160 software vs hardware (alongside a software 165)...but the thoughts here come from noticing the trend over the last few years with the various hardware units Ive had and trying the latest software that comes out.
Compressors with complex tone and non-linear characteristics I assume are very tough to model - but there is one aspect with digital compression algos that I notice aside from that and I wonder if it can be improved. This is the way the compressor clamps down and holds bass freq. In the two emulations I was comparing the higher freq response seemed almost identical eg how they responded to the higher freq drum transients. However the hardware seemed to clamp down and punch the low freq such as bass drum more easily and convincingly. The software sounded similar in the low freq but lacked the convincing gut-punching clout.
Ive heard the terminology of digital compressors 'falling apart' under strain of heavy compression which seems odd as in a digital system I dont get why a algo would feel strain...I would have thought an algo just works on the digital stream and only when its reconstituted by the converters does the signal gain its element of 'energy'...
The reason I post here is because I know Sasha is working on comps at the mo and Urs does tend to chip in and are both super brains - so I was hoping someone might shed some light on this observation...maybe I will try somewhere else if needed - but ideally Id love to hear from a developer rather than posting in a general forum...
I know this sounds like a software vs hardware - it is - but more a conceptual / theoretical / improvement point of view from someone who uses both.
Question for the developer brains...
-
- KVRAF
- 1735 posts since 28 Dec, 2007
-
- u-he
- 30225 posts since 8 Aug, 2002 from Berlin
-
- KVRAF
- Topic Starter
- 1735 posts since 28 Dec, 2007
thanks for responding Urs - so just to be clear - whilst you have put Sascha on the topic - you get what Im talking about?Urs wrote:Honestly, I don't know. That's why I hired Sascha in the first place
-
- u-he
- 30225 posts since 8 Aug, 2002 from Berlin
-
- KVRAF
- Topic Starter
- 1735 posts since 28 Dec, 2007
ah ok fair enoughUrs wrote:No, really, my expertise on compression is close to zero. I've never used an analogue one.
-
- KVRian
- 1141 posts since 2 Oct, 2001 from Berlin, Germany
I think there's a couple of reasons why analog comps can sound pretty different from one another, especially in the bass region. This isn't restricted to a digital-vs-analog comparison, moreover it affects most of the hardware comps as a whole.
One reason could be the frequency response of the detector path, aka sidechain. Most hardware comps I know or got my hands on have hp filtering somewhere on the ins & outs of the signal chain, as to remove any DC along the way. This might also apply to the sidechain. So the sheer presence of one or two caps can already alter the control element's response to LF input.
One shouldn't also forget that any filtering (at least in the analog domain, likewise IIR filters in DSP) comes with a change in phase response.
If you look at a typical signal path or schematic of a hardware compressor. you'll observe quite an amount of time-variant parts (caps, inductors, the latter being more rare here). This is true for parts that can be spotted right away, but also for what was implemented on an OP-amp's dye. A typical audio OP like a TL071 or an NE5534 has plenty of phase offset in the main signal path (be it the inverting or non-inverting branch), by means of capacitances or its transistors, let alone the internal feedback loop and measures to keep up operational params (symmetry, current limit, temperature drift etc.). Typically, several active stages (either discrete or OP-based) are interconnected, to ensure a proper level or impedance matching for surrounding parts (buffer stages, makeup amps, inverters, sum amps etc.). The resulting effects can be minor, but everythings adds up to some amount in the end.
Another aspect, apart from frequency & phase alteration, could be intermodulation products. I've heard many digital comps having difficulties with bass-heavy material and twisting the waveform beyond recognition by following the amplitude too rapidly. I suspect many algorithmic approaches being too simplistic, too 'pure', too 'text book', lacking interaction, feedback or general complexity. Call it 'soul'. I personally favour 'forgiving' algos, I tune my stuff for 'fool-proofness'. That sometimes collides with the purists among the users, but I rarely care. Some of my code is like a robot out of control, turning bad on me whenever I don't watch out, and in those moments, you get to know your beasts in a new way. This is the hardest part of design and it can takes ages to get right (like educating a dog or a small child). It just doesn't calculate well or doesn't show up in SPICE. Or, I'm just not smart enough
One reason could be the frequency response of the detector path, aka sidechain. Most hardware comps I know or got my hands on have hp filtering somewhere on the ins & outs of the signal chain, as to remove any DC along the way. This might also apply to the sidechain. So the sheer presence of one or two caps can already alter the control element's response to LF input.
One shouldn't also forget that any filtering (at least in the analog domain, likewise IIR filters in DSP) comes with a change in phase response.
If you look at a typical signal path or schematic of a hardware compressor. you'll observe quite an amount of time-variant parts (caps, inductors, the latter being more rare here). This is true for parts that can be spotted right away, but also for what was implemented on an OP-amp's dye. A typical audio OP like a TL071 or an NE5534 has plenty of phase offset in the main signal path (be it the inverting or non-inverting branch), by means of capacitances or its transistors, let alone the internal feedback loop and measures to keep up operational params (symmetry, current limit, temperature drift etc.). Typically, several active stages (either discrete or OP-based) are interconnected, to ensure a proper level or impedance matching for surrounding parts (buffer stages, makeup amps, inverters, sum amps etc.). The resulting effects can be minor, but everythings adds up to some amount in the end.
Another aspect, apart from frequency & phase alteration, could be intermodulation products. I've heard many digital comps having difficulties with bass-heavy material and twisting the waveform beyond recognition by following the amplitude too rapidly. I suspect many algorithmic approaches being too simplistic, too 'pure', too 'text book', lacking interaction, feedback or general complexity. Call it 'soul'. I personally favour 'forgiving' algos, I tune my stuff for 'fool-proofness'. That sometimes collides with the purists among the users, but I rarely care. Some of my code is like a robot out of control, turning bad on me whenever I don't watch out, and in those moments, you get to know your beasts in a new way. This is the hardest part of design and it can takes ages to get right (like educating a dog or a small child). It just doesn't calculate well or doesn't show up in SPICE. Or, I'm just not smart enough
Last edited by sascha on Sat Nov 17, 2012 9:50 pm, edited 2 times in total.
Sascha Eversmeier [formerly digitalfishphones]
TOURAGE DSP
croquesolid drum processor- mix real drums fast & focused
TOURAGE DSP
croquesolid drum processor- mix real drums fast & focused
-
- KVRAF
- 24451 posts since 7 Jan, 2009 from Croatia
Tremendous knowledge, Sascha. It's really obvious now why Urs hired you 
-
- KVRAF
- Topic Starter
- 1735 posts since 28 Dec, 2007
thanks so much for replying sascha this was exactly what I was hoping for...sascha wrote:I think there's a couple of reasons why analog comps can sound pretty different from one another, especially in the bass region. This isn't restricted to a digital-vs-analog comparison, moreover it affects most of the hardware comps as a whole.
One reason could be the frequency response of the detector path, aka sidechain. Most hardware comps I know or got my hands on have hp filtering somewhere on the ins & outs of the signal chain, as to remove any DC along the way. This might also apply to the sidechain. So the sheer presence of one or two caps can already alter the control element's response to LF input.
One shouldn't also forget that any filtering (at least in the analog domain, likewise IIR filters in DSP) comes with a change in phase response.
If you look at a typical signal path or schematic of a hardware compressor. you'll observe quite an amount of time-variant parts (caps, inductors, the latter being more rare here). This is true for parts that can be spotted right away, but also for what was implemented on an OP-amp's dye. A typical audio OP like a TL071 or an NE5534 has plenty of phase offset in the main signal path (be it the inverting or non-inverting branch), by means of capacitances or its transistors, let alone the internal feedback loop and measures to keep up operational params (symmetry, current limit, temperature drift etc.). Typically, several active stages (either discrete or OP-based) are interconnected, to ensure a proper level or impedance matching for surrounding parts (buffer stages, makeup amps, inverters, sum amps etc.). The resulting effects can be minor, but everythings adds up to some amount in the end.
Another aspect, apart from frequency & phase alteration, could be intermodulation products. I've heard many digital comps having difficulties with bass-heavy material and twisting the waveform beyond recognition by following the amplitude too rapidly. I suspect many algorithmic approaches being too simplistic, too 'pure', too 'text book', lacking interaction, feedback or general complexity. Call it 'soul'. I personally favour 'forgiving' algos, I tune my stuff for 'fool-proofness'. That sometimes collides with the purists among the users, but I rarely care. Some of my code is like a robot out of control, turning bad on me whenever I don't watch out, and in those moments, you get to know your beasts in a new way. This is the hardest part of design and it can takes ages to get right (like educating a dog or a small child). It just doesn't calculate well or doesn't show up in SPICE. Or, I'm just not smart enough
so what Im getting from this is that the difference in bass handling is not due to the ability of a digital model to process audio but the way in which they are coded...
On one hand it might be that there are aspects of the hardware that are not covered in the model such as filtering in the sidechain or small changes in phase...
Also the intermodulation products idea is very interesting...a fascinating look into the elements of coding compressors.
One thing Im always encouraged by is hearing its not a limitation of digital but hopefully can be rectified...and encouraging to hear the word 'soul'...when it comes to music...
-
- KVRAF
- 1617 posts since 11 Dec, 2008 from Minneapolis
This sounds kind of awesomesascha wrote:ome of my code is like a robot out of control, turning bad on me whenever I don't watch out, and in those moments, you get to know your beasts in a new way. This is the hardest part of design and it can takes ages to get right (like educating a dog or a small child).
Something about machine intelligence and human intelligence clashing, it's not always as straightforward as Kasparov vs. Deep Blue, and usually a bit more slapstick.