Can a hobbyist use Sharc DSP's?

[Rasz]

you can delay video instead

[BrianHG]

Oh then in that aspect it doesn't require hard real time. I do want to keep it under 40 ms so it is only one frame off sync when watching video.

What's your video source?

You know that if everything is coming from your PC, an 8 channel sound card + your own created wrapper sound driver which then feeds the hardware core audio driver will allow you to make any filters in software you like, converting any X source audio channels, be it stereo, or your own surround mixer, to any output with any filtering you like + there will be no audio/video delay since your wrapper driver tells the OS that there is an audio delay of X ms and the video playback will automatically lip-sync to the setting.

[poorchava]

You can look at TMS320 from TI. Delfino and piccolo families are rather cheap to start with, sub 200€ if memory serves me right.

But beware, documentation is fragmented as hell, they have their own names for everything you might want to search for and the processor behaves in generally very different way than your standard Cortex-M microcontroller. There is also stuff missing where you'd normally expect it (no interrupt priorities, mostly no DMA for communication interfaces)

On the other hand documentation if free, and there kinda is some help at TI community as long as you have one of the more popular processors. They are also available in packages ranging from reasonably small (they even have SOP-like if memory serves me right) to very big. Bigger members of family have CLA which is actually an independent coprocessor. The biggest Delfino I think is dual core + dual CLA all running at 200MHz. That is plenty for audio stuff.

[BCZ]

What's the advantage if delay wasn't a problem?

I can't do a PC setup, because I eventually want the DSP to be inside the speaker for a pretty solution. I started off with just a speaker, and the number of different boxes and wires was insane, something like 11 boxes and 40+ sets of cables. It was crazy messy. Now the V2 speaker has the amplifiers and power supply inside the speaker, which reduces the number of boxes by 8 and a lot of wiring, but there is still a ton of wires from the DSP's and Wi-Fi streamer. For the V3 speaker I want the DSP to be inside the speaker and completely eliminate wires and extra boxes, just a speaker and a power cord.

[b_force]

The DSP is outside the speaker, but the location doesn't matter.

I'm not sure what you guys mean. The DSP can only be used after the source and before the amplification of the speaker. I'm correcting the speaker's response, like a passive crossover, but of course I'll be doing much more than what a passive crossover can do with delays, filtering, bass boost, limiter, equal loudness contour, harmonic synthesis (fake deep bass), etc.

You can do all these things perfectly with a ADAU1701 or other sigma DSP units.
Much cheaper and less complex.

The Sharc DSPs are highly overated.
Have done quiet some PCB design  for professional audio customers. Almost all of them want a Sharc, but they barely use 10% of what it's capable of.
Unless you need a lot of memory (delays) or something very close to realtime adjustements (feedback) or monitoring, there is no reason to take a Sharc.

Sigma dsps can be easily programmed with sigmastudio with a very cheap 3rd party programming board.

[diyaudio]

The DSP is outside the speaker, but the location doesn't matter.

I'm not sure what you guys mean. The DSP can only be used after the source and before the amplification of the speaker. I'm correcting the speaker's response, like a passive crossover, but of course I'll be doing much more than what a passive crossover can do with delays, filtering, bass boost, limiter, equal loudness contour, harmonic synthesis (fake deep bass), etc.

The Sharc DSPs are highly overated.

Total Rubbish. It's the only DSP processor that I came across with SPORT channels with up to 8 channels of dedicated I2S lanes with
(Audio Decoders in ROM) It's also much simpler to program, with excellent documentation.

[b_force]

The DSP is outside the speaker, but the location doesn't matter.

I'm not sure what you guys mean. The DSP can only be used after the source and before the amplification of the speaker. I'm correcting the speaker's response, like a passive crossover, but of course I'll be doing much more than what a passive crossover can do with delays, filtering, bass boost, limiter, equal loudness contour, harmonic synthesis (fake deep bass), etc.

The Sharc DSPs are highly overated.

Total Rubbish. It's the only DSP processor that I came across with SPORT channels with up to 8 channels of dedicated I2S lanes with
(Audio Decoders in ROM) It's also much simpler to program, with excellent documentation.

That's not rubbish, you're talking about a specific solution.
Not a general audio filter solution, like the TS was mentioning.
 
Compare apples with apples, not with other fruits.   
I can name a million other projects where a Sharc DSP is an absolute necessity.
Not for just some active filtering.

[diyaudio]

The DSP is outside the speaker, but the location doesn't matter.

I'm not sure what you guys mean. The DSP can only be used after the source and before the amplification of the speaker. I'm correcting the speaker's response, like a passive crossover, but of course I'll be doing much more than what a passive crossover can do with delays, filtering, bass boost, limiter, equal loudness contour, harmonic synthesis (fake deep bass), etc.

The Sharc DSPs are highly overated.

Total Rubbish. It's the only DSP processor that I came across with SPORT channels with up to 8 channels of dedicated I2S lanes with
(Audio Decoders in ROM) It's also much simpler to program, with excellent documentation.

That's not rubbish, you're talking about a specific solution.
Not a general audio filter solution, like the TS was mentioning.
 
Compare apples with apples, not with other fruits.   
I can name a million other projects where a Sharc DSP is an absolute necessity.
Not for just some active filtering.

You keep missing the point!!. (it's obvious you haven't work with the processor) people are looking for multi-channel processing even if it's a simple IIR/FIR filter, who cares! the point is it's a multi-channel processor. i.e you can do things like (x2 3-way,  x1 center, and x1 subwoofer that's 8 channels) at 192KHz sample rate (and it won't choke), explain to me how you going to do that with other processors?

[nctnico]

Easy. Daisy chain several CODECs on an SPI bus and use DMA to transfer the data. Ofcourse the CPU should be able to handle the amount of data and algorithm but that should go without saying.

[diyaudio]

Easy. Daisy chain several CODECs on an SPI bus and use DMA to transfer the data. Ofcourse the CPU should be able to handle the amount of data and algorithm but that should go without saying.

such a cost reduction implementation!. you, of course, know I meant out of the box solution.

[ohdsp]

The ADAU1452 Sigma DSP is a very very powerful device for the price, plenty of TDM and I2S interface options (48 TDM channels in, 48 TDM channels out), built in ASRC functions and many other useful and powerful features. I have worked with professionals before who claim you can get as much, if not more, processing power from the ADAU1452 than some of the SHARC devices in the right applications as the core is based on a newer better design....

[Marco]

Unless you need a lot of memory (delays) or something very close to realtime adjustements (feedback) or monitoring, there is no reason to take a Sharc.

The popular hobby at the the moment is room correction, which requires quite long filters. Since the people doing it generally don't want or can't do much real DSP programming, they often do it with a straight FIR filter ... at that point you need a whole lot of power.

BTW, the Bela audio board has plenty of channels too.

[BCZ]

The popular hobby at the the moment is room correction, which requires quite long filters. Since the people doing it generally don't want or can't do much real DSP programming, they often do it with a straight FIR filter ... at that point you need a whole lot of power.

BTW, the Bela audio board has plenty of channels too.

Wait, you're saying there's a much more computationally efficient way to do the FIR filtering for stuff like room correction if people can do real DSP programming? Can you expand on that?

[Marco]

As I said, FFT based filtering can be used even if you want low latency (zero latency FFT based filtering).

[b_force]

The DSP is outside the speaker, but the location doesn't matter.

I'm not sure what you guys mean. The DSP can only be used after the source and before the amplification of the speaker. I'm correcting the speaker's response, like a passive crossover, but of course I'll be doing much more than what a passive crossover can do with delays, filtering, bass boost, limiter, equal loudness contour, harmonic synthesis (fake deep bass), etc.

The Sharc DSPs are highly overated.

Total Rubbish. It's the only DSP processor that I came across with SPORT channels with up to 8 channels of dedicated I2S lanes with
(Audio Decoders in ROM) It's also much simpler to program, with excellent documentation.

That's not rubbish, you're talking about a specific solution.
Not a general audio filter solution, like the TS was mentioning.
 
Compare apples with apples, not with other fruits.   
I can name a million other projects where a Sharc DSP is an absolute necessity.
Not for just some active filtering.

You keep missing the point!!. (it's obvious you haven't work with the processor) people are looking for multi-channel processing even if it's a simple IIR/FIR filter, who cares! the point is it's a multi-channel processor. i.e you can do things like (x2 3-way,  x1 center, and x1 subwoofer that's 8 channels) at 192KHz sample rate (and it won't choke), explain to me how you going to do that with other processors?

I am not missing the point, but you're only pointing out one specific market.
What you're referring to, is only suitable for home/cinema audio type like market.
There are many different areas where people don't even care about multi-channel audio.

The AUDAU1701 etc has 4 channels out, or can even be used with an high quality DAC, CODEC and ADC.
More than good enough for most active mutli-way speaker systems (unless you're looking for something specific)


Yes, I have worked with these processors quiet a bit.
BTW, I am not saying that the Sharc is rubbish, I am just saying that it's heavily overrated in just fairly simple active filter designs.
The topic starter (BCZ) clearly said he just wants to make some active filters with it.

If it's just for hobby, I would recommend building a dedicated media center pc instead.
Much cheaper, much more possibilities, virtually endless delays.

[b_force]

Unless you need a lot of memory (delays) or something very close to realtime adjustements (feedback) or monitoring, there is no reason to take a Sharc.

The popular hobby at the the moment is room correction, which requires quite long filters. Since the people doing it generally don't want or can't do much real DSP programming, they often do it with a straight FIR filter ... at that point you need a whole lot of power.

BTW, the Bela audio board has plenty of channels too.

Room correction is not just for hobby or 'popular', it's a fundamental principal in room acoustics.
Widely used in professional setups for many years.
What you're referring to is delay correction with a multi-subwoofer system.

For what BCZ is planning to do with it, you don't need a Sharc and you definitely don't need FIR filters.

[diyaudio]

Unless you need a lot of memory (delays) or something very close to realtime adjustements (feedback) or monitoring, there is no reason to take a Sharc.

The popular hobby at the the moment is room correction, which requires quite long filters. Since the people doing it generally don't want or can't do much real DSP programming, they often do it with a straight FIR filter ... at that point you need a whole lot of power.

BTW, the Bela audio board has plenty of channels too.

Room correction is not just for hobby or 'popular', it's a fundamental principal in room acoustics.

We in audiophoolery territory. 

[b_force]

Unless you need a lot of memory (delays) or something very close to realtime adjustements (feedback) or monitoring, there is no reason to take a Sharc.

The popular hobby at the the moment is room correction, which requires quite long filters. Since the people doing it generally don't want or can't do much real DSP programming, they often do it with a straight FIR filter ... at that point you need a whole lot of power.

BTW, the Bela audio board has plenty of channels too.

Room correction is not just for hobby or 'popular', it's a fundamental principal in room acoustics.

We in audiophoolery territory. 

Go read some stuff on Floyd Toole, John Eargle and basic acoustics.
Or just call any acoustic consultant company (you know, the ones who develop concert halls, churches, theaters etc)
Nothing audiophile about this stuff. Basic physics.

[Marco]

The topic starter (BCZ) clearly said he just wants to make some active filters with it.

I'm not familiar with exactly how it's implemented, but I don't see how the ADAU's can do base harmonic synthesis at least. He'll need something properly programmable with some significant amounts of memory for that.

Room correction is not just for hobby or 'popular'

That doesn't mean you can't do it as a hobby. Speaker design was always a common hobby, a hobby which has entered the digital realm for a while already.

[Rasz]

I get it now. OP wants DSP because its modern wooden volume knob/oxygen free gold plated power cable, goes to 11 etc.


How about some acoustic panels on your ceiling/walls?

[BCZ]

OK I want to step in and explain what I want in much more detail because even on DIY speaker forums few people understand the complexity of what I want to do. It is clear there are people here who are very knowledgeable in audio.

First, if I only need to do a couple of active filters, I would just stick with a miniDSP and call it a day. Boom, done, simple, small, and not horribly expensive. Of course, I need the common active filters like Linkwitz Riley filters for crossover and shelf filters for adjusting the frequency response for 8 channels, etc. But I also need a lot more.

I want to build the best possible small speaker. There are a lot more that goes into that than just some active filters. I need "dynamic" filters.

Here is what I need that the miniDSP can't do.

- Multi-band compressor. Huge amounts of bass boost will be used. In order to protect the speaker from blowing itself up, a compressor/limiter is needed to dial back the bass when it exceeds the limits of the speaker. A compressor will compress the entire band, but I only want the compressor to compress the bass band, and hence the need of a multi-band compressor. This is easily done if a compressor is already available. Then, all that is needed is to separate the signal into multiple bands like a crossover does, apply a different compressor setting to each band, then sum everything back together.

- Equal loudness contour: At low volumes, we need a lot more bass for the sound to sound "balanced". At high volumes, our hearing sensitivity to bass and treble has increased, and we want progressively less bass and treble as volume is increased. I want to implement this to preserve tonal balance at all volume levels.

- Fake bass synthesizer. There is a psychoacoustic phenomenon where the ear can be tricked to think it is hearing a fundamental note lower than what it actually is with the right harmonic composition. It is used to make small speakers seem like it is playing bass when it is not. I want to have that for my speaker. A twist will be to activate this feature only above certain levels. For example, my speaker is capable of quite low and quite loud. It is capable of playing 40Hz at 95dB. But it can play 100Hz at 110dB. So eventually I want to program it in a way where the fake bass synthesizer is not activated until the incoming signal exceeds the limits of the speaker. For example, if I need to play 40Hz at 90dB, it'll just play 40Hz at 90dB. But if it needs to play 40Hz at 100dB, then the woofer will play 40Hz at its limit of 95dB and get the other the 5dB from the fake bass synthesizer making harmonics at 80Hz, 120Hz, 160Hz, etc where it is capable of playing 110dB.

- Lots of FIR filtering - Not only for linear phase crossovers, but also to correct group delay of woofers and room correction. For low frequencies a lot of power is needed for those long FIR's.

- Thermal compression compensator - Because the speaker is so small, the woofer requires 3-10x more power to achieve the same volume for bass as a normal sized speaker. 99% of the power is converted into heat. The problem is that the hotter the woofer gets, the more power the woofer needs to achieve the same volume because the sensitivity is reduced due to the higher temperature causing higher resistance. This means I need something to apply a boost to the woofer depending on the volume level. This is an advanced feature that needs to be written myself. A simplified implementation can be the same as the equal loudness contour, but after a certain volume it start boosting instead of cutting, but it is a "dumb" variable boost since it is not varying depending on the woofer's temperature based on a temperature model of the woofer.

- Thermal limiter: My speaker can play 105dB without distortion, but it doesn't mean it can play 105dB for long periods of time because it can't handle that level of power for very long. Therefore, the volume needs to be gradually backed down. It can play 100dB for maybe a few minutes, then it'll get too hot and need volume to be further reduced. I need a feature to monitor the average RMS level and gradually adjust the limiter up or down. So if I want to play the speaker for hours on end, it'll automatically limit the volume to say 95dB to prevent the speaker from melting. This is another feature that needs to be written myself.

I hope this gives a better idea of what I need.

[BCZ]

I forgot to say, for the first step into the foray of DSP's, I would be happy if I can just replicate the miniDSP to do the normal active filters, crossovers, time delay, FIR, and compressor for 8 channels. Just that alone would reduce the number of DSPs from 4 miniDSP's to 1 and a huge reduction in wirings. I'm using multiple miniDSP's for the sole purpose of the multi band compressor. The miniDSP has 4 channel outputs, but only a single compressor per output. Therefore I'm making a multi band compressor by summing together the outputs using analog components, which is a very silly and expensive way when all it takes is a simple addition of outputs done within the DSP and then output to one channel. But that's the crap I have to deal with miniDSP that pushed me to go with my own DSP.

All the other fancy dynamic adjustments of bass, treble, fake bass synthesizer, thermal compensation and limiter can come later.

I repeat, I cannot use a PC. I started this project because I travel A LOT (mostly not for vacations unfortunately), and I want a nice sounding small speaker to bring with me to my travels. Of course I went batshit crazy in terms of how far I wanted to push this (the speaker cost me  ~$4000 to build...), but it has been so much fun. Trust me, no audiophool BS, all science. I'm actually in the process of "upgrading" my cables from $1 cables to $10 cables because the ultra cheap cables have been causing me some problems. Hence, the small speaker and the desire to have everything in one box.

[kilohercas]

What i did is made programmer out of FT2232D, that simply disconnects boot flash from DSP, loads data from hex file, and that it, debugging is done via uart.

This is my Mandelbrot test, and ti bowed my mind how much performance where is compared to STM32F4, at least 10 times faster

[Marco]

This is my Mandelbrot test, and ti bowed my mind how much performance where is compared to STM32F4, at least 10 times faster

The M7 would be a better comparison. Harvard architecture memory banks with DMA, dual issue arithmetic/memory-ops. It's almost a DSP at this point, except like most processor designers the ARM designers can't get it through their bloody heads that you can cheaply implement something other than either brain dead branch behaviour or too smart for it's own good branch prediction. Still, it will handle branches a lot better than the M4.

[poorchava]

I wonder how does CM7s TCM improve performance over CM4s pure load-store model. I remember a project where we found out that when performing relatively simple operation on a large amount of data, memory load/store oeprations were the bottleneck of the whole system. In comparison TMS320 Piccolo was much faster as simlar FCPU.