You are right about the DSP. I have been looking for DSP powered interfaces, and they have all had eq, filters, compressors and limiter, but I see that RME is different.
Me too, and DA and wordclock... it's one of the reason i purchased a Lake for my main system. But like for you it ends up in a mess of cable and multiple box. Once you go multi amp even with multipair it's a vain attempt! At least girlfriend and other regular people are frightened to power up the whole thingy, it kind of secure the system!!!
Yes much more 'pro' oriented that other mid priced gear. One gear one function but done well and stable as a rock. Once you have some you forget about it... It's as good as this.
Last edited:
I use Xonar U7 - quite another price
The RME is a weird choise for a multichannel DAC. It has onboard DSP, so you could actually do the eq and filtering with the onboard DSP, but not FIR I guess.
There are many very good USB I/O boxes with no DSP.
Try look at the ESI Gïgaport HD+ - a no nonsense 8 output device with RCA outputs
The ESI Gigaport HD+ does "only" 6 channel at 48Khz/24 Bit. 8 Channels i limited to 41Khz/16 bit.
I have it, and I REALLY like it. It is VERY stable. Not even at single time have a seen the driver mess up with my system (I have had it for abouth two months) For my use it has more than enough output "strength (it is only driven by USB bus power). Only a little hizz when driving to maximum volume - I never have the volume in the software turned to maximum
In comparison the Focusrite Scarlett 18/20 has way to much noise for my liking. After one day I handed it back to the seller. Not a shame, because it is considerably more expensive.
One thing I would like to have have is an good analog in / digital in - Guess that can be fixed with another soundcard.
thanks. I love my adi 8s. i leave them powered on 24/7 no problems.
a quiet pc is really simple actually. if you are using it dedicated for convolution you dont need all the cd drives, etc.
so you can use a $50 fanless powersupply, and a $50 sdd drive, and either a very quiet cpu fan or even a fanless cooler. done.
As far as computer config. Also really easy(once youve done it a time or two)
1. install linux(I use a command line only version of debian 8 jessie. boots up in 23 secs from button push till im hearing audio out of it.
2 install brutefir from the software repository using this command
"apt-get install brutefir"
almost too simple
3.setup the computer to automatically load brutefir on startup.
4. use rephase to create filters and load it into the brutefir config file.
its really fairly straightforward.
the cheapest way I know for 8 channels in and 8 channels out would be the minidsp adat usbstreamer b ($105)with a behringer 8200($200)
Yes it seems to be easy and straightforward.
Which processor do you use and what is the latency (typical, it will vary with filter type and fir processing you do but this is to have a rough idea of what to expect).
I think you've got me convinced to try this! (And probably a win based solution using a vst host and Voxengo pristine space vst for deconvolution and my good old Aardvark as converters -which isn't supported under linux- for my future late night session system. Thanks.
Which processor do you use and what is the latency (typical, it will vary with filter type and fir processing you do but this is to have a rough idea of what to expect).
I think you've got me convinced to try this! (And probably a win based solution using a vst host and Voxengo pristine space vst for deconvolution and my good old Aardvark as converters -which isn't supported under linux- for my future late night session system. Thanks.
I will try to explain but you will probably find clearer and more throughout explanations on the web.
Time based convolution is the most straightforward way of doing a convolution: you basically multiply/add each sample you want to process with each sample of the FIR.
This is of course very CPU intensive but fits well with the way DSPs work, and this is the technique used in hardware devices like the HD2, openDRC, 2x4HD, najda, etc...
The advantage is that it is simple (ie implementation errors are unlikely), the result suffers almost no quantization error (ie noise), and the process does not imply any additional delay beside the one "embedded" in the FIR (the one rephase indicates when the FIR is done).
You can basically do zero-delay (beside DAC) minimum-phase FIR corrections with this kind of convolution, just like IIR.
On the other hand frequency-based convolution relies on FFTs/IFFTs of blocks of the input signal, and implies an additional delay to get that block.
Partitioning and other fancy technics help reducing this delay but add complexity.
This technique is much less straightforward and the quality of the convolution (noise, or even discontinuities between blocks) will heavily depend on the implementation. It requires more memory but much less CPU and is typically the way PC software like BruteFIR, SoX or convolver work.
I personally really like the brutally elegant and simple ways of direct time-domain convolution
or as I was taught back in college decades ago, multiplication in the frequency domain is equivalent to convolution in the time domain and that savings in computation was the primary motivation for fast (FFT) methods of transforming between domains. Progress according to Moore's law over those decades enable the simpler time domain solution.
re' staggeringly inefficient
Just harness the vector supercomputer embedded in your PC's GPU. (darn, those PC's are insidious)
It is algorithmically inefficient alright, but beautiful in its simplicity and elegance.
And the good news is that a tiny/cheap/energy-efficient DSP chip can do it just fine without breaking a sweat.
We just need to wait for a manufacturer to combine enough of these cheapy chips in one device now
And the good news is that a tiny/cheap/energy-efficient DSP chip can do it just fine without breaking a sweat.
We just need to wait for a manufacturer to combine enough of these cheapy chips in one device now
How much is enough ?
Run some sims on a sub , make a 40hz high Q linear crossover , remove some excess GD from 20-100hz and some eq with a rectangle window. Keep adding taps until sim is perfect
Whatever that number is triple it for rest of system . That should be about right
If I were to guess , 40k would probably get close.
Pos?
Run some sims on a sub , make a 40hz high Q linear crossover , remove some excess GD from 20-100hz and some eq with a rectangle window. Keep adding taps until sim is perfect
Whatever that number is triple it for rest of system . That should be about right
If I were to guess , 40k would probably get close.
Pos?
6144 taps per channel at 48kHz is just enough for my current needs.
I would love to have something like 32k taps per channel at 96kHz.
I don't know if a single DSP chip would be powerful enough to do a single channel, as it would need to be around four times the power of the 400MHz sharc used in the openDRC.
I don't see much progress in the more recent sharc DSP, beside new features and memory.
Oh well, 16k taps at 48kHz would be enough already, and should be feasible with one sharc DSP per channel.
I would love to have something like 32k taps per channel at 96kHz.
I don't know if a single DSP chip would be powerful enough to do a single channel, as it would need to be around four times the power of the 400MHz sharc used in the openDRC.
I don't see much progress in the more recent sharc DSP, beside new features and memory.
Oh well, 16k taps at 48kHz would be enough already, and should be feasible with one sharc DSP per channel.
Yes sir !!!!!
No matter how I try, I can't stretch 6144 taps to handle the mid driver the way I'd like to use it. I need 100 Hz high pass, and 650 Hz LP. The high pass needs to be as steep as possible, up to 96 db/oct. LP is no problem to implement. Sub low pass at 100 Hz no real problem (sub gets dicey with any kind of HP though
If anybody can point me to a filter type and windowing that gives best chance for the mid described above, at holding linear phase down into sub land (60-70Hz) ...please advise. thx Mark
Otherwise I'm waiting on the box POS proposes ...aargh
No matter how I try, I can't stretch 6144 taps to handle the mid driver the way I'd like to use it. I need 100 Hz high pass, and 650 Hz LP. The high pass needs to be as steep as possible, up to 96 db/oct. LP is no problem to implement. Sub low pass at 100 Hz no real problem (sub gets dicey with any kind of HP though
If anybody can point me to a filter type and windowing that gives best chance for the mid described above, at holding linear phase down into sub land (60-70Hz) ...please advise. thx Mark
Otherwise I'm waiting on the box POS proposes ...aargh
use multiple open drc boxes. Solved
I use a 1in 3out toslink splitter it works fine on the 3-HDs , imagine what 3 or 4 opendrc boxes could accomplish. Pay to play I guess but I don't see any other way to get a very very good correction using hardware box...could use separate dacs and everything....
Who's going to go 1st?
I use a 1in 3out toslink splitter it works fine on the 3-HDs , imagine what 3 or 4 opendrc boxes could accomplish. Pay to play I guess but I don't see any other way to get a very very good correction using hardware box...could use separate dacs and everything....
Who's going to go 1st?
Hi, I have 3 open-drc's fed with a spdif coax splitter...
but how does that solve getting more than 6144 taps on any one bandpass filter ?
......ie the 100Hz to 650 HZ bandpass I'm having trouble trying to get a steep enough HP, and still maintain flat phase....?
Nice setup!
Well in your case maybe you would have to add another box and series it with your midbass drivers , do two LR4 (or whatever) filters cascaded acrosss two units. Would be double the delay and would need something downstream that can facilitate that much delay across channels.... just an idea
Don't know what dacs you use but I assume you have the delay thing figured out
I used two cascaded filters out of one HD and looped back into it (yes had one ADC in the mix) it wasn't bad for sub and it worked ...I could only get 2042 taps per side so I made two 2042 tap filters and ran them into each other...it worked, for the open drc tho you won't need use ADC like I had to do it should be fine
Well in your case maybe you would have to add another box and series it with your midbass drivers , do two LR4 (or whatever) filters cascaded acrosss two units. Would be double the delay and would need something downstream that can facilitate that much delay across channels.... just an idea
Don't know what dacs you use but I assume you have the delay thing figured out
I used two cascaded filters out of one HD and looped back into it (yes had one ADC in the mix) it wasn't bad for sub and it worked ...I could only get 2042 taps per side so I made two 2042 tap filters and ran them into each other...it worked, for the open drc tho you won't need use ADC like I had to do it should be fine
Last edited:
you should try it. the hardware boxes dont come close in functionality.
Im using a 3.2ghz processor. I have it set to 256 samples brutefir delay and Im using 8192 taps with a 5% impulse center(409.6 samples) so total delay is 665 samples or 15ms
I just logged in remotely
(yeah, cant do that with an opendrc) and checked the processor and memory usage using the command "top"hardly even breaking a sweat at 12.7% cpu usage and 3% memory usage.
this is with 10 channels/filters at 8192 taps each.
for the record, if there are no lipsync needs for home theater or live sound, you can run brutefir with no partitions (longer system delay)and it will happily run at 0-2% cpu all day
if you are using windows 10 keep in mind many older soundcards will not work with windows 10.
I use a 9636 and two rme adi 8 converters which means I have 16 inputs and 16 outputs available to me. for my front three speakers which are three way active I use 9 channels and channel 10 is for LFE.
