Author Topic: OT digital rack unit or digital software  (Read 3740 times)

1wahfreak

OT digital rack unit or digital software
« on: December 27, 2004, 05:39:31 PM »
So I'm wondering if there is any real difference bewtween a rack mount digital reverb vs digital software. I know there are different levels of each, but let's assume Lexicon has there own version of rack and software(which they do). Rack units use their own internal processor vs software which uses your computers processor. Is the fidelity based on processor capabilities?
Opinions??

MartyMart

OT digital rack unit or digital software
« Reply #1 on: December 27, 2004, 07:11:52 PM »
I use both in the studio, software versions are now very much up to the job, however I find that a "rack" unit, such as a Lexicon reverb, is still somehow easier to "sit" in a mix than a sofware version.
Remember that a good reverb will "eat" your computers porcessor up very quickly indeed, so you will have to set one up as a send/return, just like a real desk and patch your recorded tracks to it in the same way,
"Altiverb" and Emagic's "Space designer" are fantastic reverb plugins.

Marty.
"Success is the ability to go from one failure to another with no loss of enthusiasm"
My Website www.martinlister.com

Paul Perry (Frostwave)

OT digital rack unit or digital software
« Reply #2 on: December 27, 2004, 08:20:22 PM »
A dedicated unit is potentially superior, because it could be optimised for noise considerations, it doesn't have problems wiht other computer tasks fighting for resources, might have a dedicated audio DSP chip. Plus you don't lose an in/out (or two!) on your PC/audio interface when you link to otther FX.
But, it might not either.......and it is going to cost..

for me, the biggest problem with a PC solution is that you can't stick your favorite FX boxes into the PC! :x

SeanCostello

OT digital rack unit or digital software
« Reply #3 on: December 28, 2004, 02:02:16 AM »
Many dedicated rack reverbs will use fixed point processors. The standard processor is the Motorola 56k series, which works in 24 bit. So, the audio quality of a rack processor might be somewhat less than the 32 bit floating point used in most PC/Mac plugins.

But...

The fixed point processors might use double precision, which results in 48 bit fixed point, which is far superior to 32 bit float.

And...

Some rack processors use SHARC DSPs from Analog Devices, which are 32 bit floating point processors, and can perform 64 bit double precision fixed point as well. So, the audio quality of these would be higher than 32 bit floating point on a PC/Mac.

However...

You can also program plugins on a PC or Mac in double precision. Which can be up to 80 bit, if I recall correctly. It will slow things down, but it will have good numeric properties. Basically, this matters when you have filters with low cutoff frequencies, or filters with high feedback gains - which reverbs have.

In the past few years, there has been a lot of discussion about how native processing (i.e. PC/Mac) can be superior to the DSP/FPGA/custom silicon solutions found in rackmount gear. However, when you start actually programming plugins, you find out that there are some major limitations of a PC versus a dedicated DSP (I should note here that I program DSPs for Analog Devices, so take what I say with a grain of salt, but I was programming on PCs before I programmed DSPs). DSPs are designed to crunch numbers repetitively, and they do so very well. I programmed a VST spring reverb simulator this summer, that used 70% of a 2.66 GHz Pentium 4. The same algorithm took up about 8% of a 200 MHz SHARC. Mind you, I probably programmed the algorithm more efficiently on the SHARC (I would work with a block of samples at a time on the SHARC, while the VST framework I was working with calculated a sample at a time), but this is still a very significant difference. In general, I have found that many of the algorithms that would make my PC choke work well on the SHARC, even though the PC is much faster in terms of MHz. I suspect that the same is true of Motorola and TI processors, although those processors are well known for starting fires and killing puppies.  :wink:

The main thing about reverbs is memory access. Reverbs demand fast memory access, in order to implement as many delay taps as they can get away with. David Griesinger of Lexicon has said in the past that he can create a good reverb with 100 taps - but this is still a lot of memory accesses. Current reverb designs, such as the TC Electronic designs and the high end Lexicons, probably use on the order of 400 to 600 taps (and don't ask me how these taps are arranged, as I don't know and don't want to know, as it would present IP challenges to designing my own reverb designs). In order to access these taps, you need fast memory access, as well as the DSP horsepower to perform the circular buffering, which entails checking for the wraparound of the end of the buffer. Some reverbs, such as the Quantec Yardstick and TC Electronics reverbs, use FPGA chips. It is likely that these chips are used to accelerate the memory access, and possibly perform some of the circular buffering functions.

The non-convolution reverbs I have seen for PCs/Macs seem to use a lot less taps than the rackmount reverbs. This may be because it is expected that multiple instances of the reverb will be run on the PC, together with other effects and processes. Still, my feeling is that most plugins will pale in comparison to even the medium-end rackmount reverbs from TC Electronics and Lexicon.

Another factor: Getting low latency in a rackmount processor is simply a matter of design. Use low-latency A/D and D/A convertors, choose a suitable block size for the processing, and you have low latency. No worries about any other processes interrupting the reverb.

I have to admit that the convolution reverbs such as Altiverb and Space Designer sound good, and work well on PCs and Macs (especially Power PC chips that have good SIMD FFT routines). Still, this is sort of a "brute force" approach to reverb. The fact is, a good algorithm can use a small fraction of the resources of another reverb algorithm, and sound much better. Lexicon had great sounding reverbs in the late 1970's, with almost no RAM or DSP cycles available. In this case, a good reverb algorithm will sound good on PCs as well as dedicated hardware.

One other thing to note: Convolution reverbs will not be able to obtain the "time-varying" reverbs made famous by Lexicon, Ursa Major and others. Having the reverb delay and gain coefficients slowly varying over time will give you a sound that avoids metallic coloration, and can sit really nicely in a mix. A convolution impulse of a Lexicon 480 will only capture a snapshot of this evolving sound.

OK, enough of my rantings on this subject.

Sean Costello