DSP Info

[BKr0n]

For those of you just joining us: this thread is for covering some info about DSP.

Pertaining to me: I have been working on making a set of bookshelf for a little bit now. One thing I've been waffling back and forth on is whether to go DSP or full analog. One thing I've ran into a lot is the concept of real time systems. I've discovered that there is far more to this than just the simple term "real time". So much so in fact, it has sort of become its own field of study with quite a bit of granularity, but I digress.

I guess for starters here: what are two examples of both ends of the spectrum when it comes to DSP in audio (like from simplest to most complex)? (@kemmler3D batter up )

 

[kemmler3D]

Hmm, interesting way to phrase the question...

I'd say the simplest form of DSP is a tone knob (bass/treble) on a digitally enabled DSP or amp.

The most complex form of DSP (in terms of difficulty to implement) would probably be like what you're talking about with realtime, and from scratch, i.e. coding your own DSP algorithms in a low level language.

In general using a true realtime OS is not required for audio since most DSP units can achieve 20ms of latency, often way less, which is fast enough that we don't really perceive a delay anyway. The exception would be if you're doing servo control or something that relies on ~zero latency.

My questions - what are you trying to achieve with the bookshelves, what have you tried so far, what are your design constraints, and what's your budget? I gather you have a high tolerance for tinkering, which is good.

 

[BKr0n]

what are you trying to achieve with the bookshelves

I would like to have a set of bookshelves for a small to medium size room with as little distortion as possible, can get to a reasonable loudness, and only be forward firing. I originally started this because I got a decent deal on a set of KEF SP1632s, but the funny part is I may not even use them in this project. It uh... kind of keeps evolving lol

what have you tried so far

So, this starts with the caveat: I do have some engineering background, but did not have as much geared toward audio. This has definitely been a learning experience.

It started with the SP1632s. I figured sticking to a compact design go with a compact woofer. The initial design was with a dayton 5 inch epique woofer. What i didn't realize was all the foibles that would entail. The main one being the SP1632 has a sensitivity of 93db, while the dayton only has that of 83db. After a little while I got tired of fighting physics and started looking at other drivers. Honestly? At this point I'm thinking of just re-selecting drivers from the ground up. With what I've learned so far it may behoove me to do so.

The other thing I've been back and forth on is analog or digital. Analog, for the most part I've found to be quite a bit easier (input section for EMI, crossover, and LR transform, and main amp to drive the speakers). Digital, however, I realize has far more flexibility on all of those fronts, and can even add to it (things like FFTs, small signal analysis, etc). But I also came to realize that the more I wanted to do with it, the more complex it became. That goes double for the quality of components ( ex: using something like an ESS9038PRO vs a PCM1748). When I started getting to Sharc processors I started to realize I was going overboard and said maaaaaaybe just go analog.

what are your design constraints, and what's your budget?

At this point, I would be willing to dip into my savings a little bit, but I would need to be able to justify the expenditure. I at least have enough background to know what I'm reading on a spec sheet

I gather you have a high tolerance for tinkering, which is good.

Have tinkered both personally and professionally. I've worked on all kinds of stuff from consumer to aerospace and industrial. I would like to think I definitely have a good foundation for getting into DIY audio.

 

[kemmler3D]

a set of bookshelves for a small to medium size room with as little distortion as possible, can get to a reasonable loudness, and only be forward firing.

Definitely all doable, depending on how little distortion you really want.

Purifi drivers have the lowest distortion I've seen so far, but they're $$$, maybe more than the KEFs.

In any case, the drivers will introduce vastly more distortion than the DAC or amp, so I would focus your attention there.

In general, the way to design a speaker starts with deciding on:

• Frequency range you want
• SPL you want
• Directivity you want
• Design paradigm you want (OB, box, cardioid, horn, etc.)
• Size you're willing to live with
• $$$ you're willing to spend

From there, you can decide on drivers, the drivers dictate the box, (waveguide maybe) and crossover, and then it's a matter of measuring, tweaking, and putting it all together.

"Only forward firing" needs a little clarification. Most speakers that aren't OB / dipole / omni are nominally forward-firing. Then you have the cardioid speakers like D&D or Sigberg that are actually more forward-firing, across more of the spectrum, than most box speakers. The latter is said to sound amazing, but it's much more complex to build.

Then you have the related question of how wide of dispersion you want. Do you want a wide "sweet spot" at the expense of more sound bouncing off the walls, or do you want narrow dispersion, with perhaps more control over tonality, but at the cost of needing to sit relatively close and not move around too much?

I think there are decent arguments for going with analog vs. digital. In my personal opinion, the big (BIG) advantage of digital over analog is you don't have to physically build and rebuild the crossover, you also have much more flexibility for correcting outside of the crossover region if you need to. With enough amp power you can also do a lot to extend / fill in the bass with DSP and take your drivers right up to their limits.

But, as you note, it adds more components, potentially an ADC section, and there's a whole range of options to deal with.

This is why I keep bringing up the hypex amps. Bottom line, you get 1-3 amps, reasonably good DSP, good ADC / DAC section, and it's all in one tidy package.

SHARC stuff is well and good but I think the main reason to go that route instead of already-built plate amps is to save money, or if you have some notion of programming a custom UI, stuff like that.

 

[BKr0n]

In any case, the drivers will introduce vastly more distortion than the DAC or amp, so I would focus your attention there.

Yeah that seems to be another thing I keep running into. I think it's rather interesting how so many amplifier designs cut the speaker completely out of the equation when that's going to be the weakest link in the chain.

Purifi drivers have the lowest distortion I've seen so far, but they're $$$, maybe more than the KEFs.

Yes they are... I know because I was looking at them next lol

"Only forward firing" needs a little clarification.

So I've noticed that some speakers (actually D&D is a good example) that they have either PRs or speakers on the backs of them. Where these are going they won't have much of a wall behind them to deflect off of.

Then you have the related question of how wide of dispersion you want. Do you want a wide "sweet spot" at the expense of more sound bouncing off the walls, or do you want narrow dispersion, with perhaps more control over tonality, but at the cost of needing to sit relatively close and not move around too much?

The overarching plan is to have these hook up to something like a topping dac and then have the dac hook right into my computer. I figured that, even if I did go dsp, it would take a majority of the processing load off of the speakers so that their only job is sounding good. That would also mean if somewhere down the line I wanted to use correction software, it would be far easier with my PC than if I put a dedicated piece of silicon in the speakers themselves.

This is why I keep bringing up the hypex amps. Bottom line, you get 1-3 amps, reasonably good DSP, good ADC / DAC section, and it's all in one tidy package.

SHARC stuff is well and good but I think the main reason to go that route instead of already-built plate amps is to save money, or if you have some notion of programming a custom UI, stuff like that.

I guess it really all just comes down to the quality of the signal. Whichever is cleaner is the one I'll go with. With all that I've read, I'll admit, I definitely should take a closer look. are there any others out there besides hypex, or is that about as good as it gets in terms of DSP amps you can put in speakers?

Oh, and as far as Sharc goes, sigmastudio is a REALLY powerful piece of software. There's even dev boards specifically for that use (https://www.analog.com/en/design-ce...valuation-boards-kits/sharc-audio-module.html). But as previously stated, complexity got to the point where I had to kind of tug on my own reigns.

One thing I have noticed though with active DSP is that it mainly only sticks to the 48khz range for sampling. Why is that?

 

[kemmler3D]

So I've noticed that some speakers (actually D&D is a good example) that they have either PRs or speakers on the backs of them. Where these are going they won't have much of a wall behind them to deflect off of.

Yep. However, they are not relying on the wall for the most part. The rear radiators are used to cancel out sound emanating from the front, so it doesn't have a chance to bounce off the walls. The key thing to remember here is that at low frequencies the wavelengths are really long, such that the ~100hz waves are much bigger than the entire speaker, and are expanding more or less omnidirectionally. So they come from the front, but also propagate toward the back - where the rear-facing drivers cancel them out and create the front-facing "cardioid" radiation pattern.

That's where the DSP comes in. You need to attenuate the sound coming from the rear exactly the right amount, and delay it exactly right, so that it cancels out and doesn't just add another bouncing sound wave into the room.

I figured that, even if I did go dsp, it would take a majority of the processing load off of the speakers so that their only job is sounding good. That would also mean if somewhere down the line I wanted to use correction software, it would be far easier with my PC than if I put a dedicated piece of silicon in the speakers themselves.

What's common is to use the DSP in the speaker to correct the speaker independently of the room. Then, you can run room correction on the PC.

You CAN do room correction in the speaker itself, too.

In general DSP of this type is not very demanding on the PC or the digital bits in the speaker either. It's been pretty easy for a computer to do for 20+ years. You should decide how many boxes and wires you want to deal with, and how you'd like to edit the various DSP settings. That's more important than how much computational load is placed on the PC or speaker.

are there any others out there besides hypex, or is that about as good as it gets in terms of DSP amps you can put in speakers?

There are Pascal amps which are supposed to be about as good, but they don't sell to consumers. I'm not aware of any other really SOTA-approaching hi-fi plate amps with good DSP.

One thing I have noticed though with active DSP is that it mainly only sticks to the 48khz range for sampling. Why is that?

I think the Hypex ones actually run at some weird sampling rate like 97khz... but in general, 48khz is good enough to reproduce any audible frequency, it's compatible with everything, and it's not demanding on the CPU.

DSP at 44.1 or 48khz is easy - 96khz is twice as hard. 384khz and you're suddenly overloading your CPU.

What a lot of people end up doing is using one or more MiniDSP units and then hooking them to outboard amps. This leads to a whole mess of wires outside the speaker, but gives you the best of both worlds in terms of high-quality components and the power of a DSP speaker.

Personally I can't tolerate any more wires in my already wire-ridden space so I'd personally go with plate amps.

My broadest advice would be to sit down and decide what kind of output you want from the speakers other than minimum distortion. If lowest-possible distortion is the only and most important goal, then you definitely go with Purifi drivers, you go with the 8" and 10" aluminum drivers, and a 4" mid, and a Bliesma tweeter, and you probably take out a mortgage to pay for them. Even with those quality of drivers, a 90 SINAD amp will still be swamped by driver distortion.

 

[BKr0n]

Yep. However, they are not relying on the wall for the most part. The rear radiators are used to cancel out sound emanating from the front, so it doesn't have a chance to bounce off the walls. The key thing to remember here is that at low frequencies the wavelengths are really long, such that the ~100hz waves are much bigger than the entire speaker, and are expanding more or less omnidirectionally. So they come from the front, but also propagate toward the back - where the rear-facing drivers cancel them out and create the front-facing "cardioid" radiation pattern.

Ah OK. From a lot of what I've been reading, most people put them on the sides to cancel out the bass out without losing directivity.

There are Pascal amps which are supposed to be about as good, but they don't sell to consumers. I'm not aware of any other really SOTA-approaching hi-fi plate amps with good DSP.

DSP-A module - Pascal Audio

www.pascal-audio.com

They do make this so maybe? I'd have to Google around.

DSP at 44.1 or 48khz is easy - 96khz is twice as hard. 384khz and you're suddenly overloading your CPU.

So I would assume then it's far easier to process longer bit words (24 and 32) as opposed to higher sample rates?

What a lot of people end up doing is using one or more MiniDSP units and then hooking them to outboard amps. This leads to a whole mess of wires outside the speaker, but gives you the best of both worlds in terms of high-quality components and the power of a DSP speaker.

I just recently saw their new htx model. Thats actually what got me wondering about 48khz. Say I use the topping dac as my source. Wouldn't there be loss, or even audible artifacts, going from analog to digital and vice versa so many times?

If lowest-possible distortion is the only and most important goal, then you definitely go with Purifi drivers, you go with the 8" and 10" aluminum drivers, and a 4" mid, and a Bliesma tweeter, and you probably take out a mortgage to pay for them.

I saw that RAAL makes a mid tweeter... ngl tempted to make that my mid and cry over my wallet after. Also, why Beliesma specifically?

EDIT: Also, why specifically 8" or 10" ? They make an aluminum that's also a 6.5" so wouldn't that be better to cut down on size?

 

[voodooless]

So I would assume then it's far easier to process longer bit words (24 and 32) as opposed to higher sample rates?

Not in all cases. Generally a DSP has one or two native bit depth it can process at (either fixed or floating point). Anything else will add significant overhead. Sample rate just means you need more cycles for the same amount of sample time, meaning that less cycles are left for more elaborate processing.

So generally, if you double the sample rate, you half the potential processing.

 

[BKr0n]

So realistically that Sharc audio board would be all I would need to put my own DSP design in?

 

[voodooless]

So realistically that Sharc audio board would be all I would need to put my own DSP design in?

That will depend on what you plan to do. If you need thousands of convolution taps, it’s not going to be a good fit.

 

[IAtaman]

Why not a cheap SBC between your source and your DAC that is running a lightweight linux distro and CamillaDSP?

 

[voodooless]

Say I use the topping dac as my source. Wouldn't there be loss, or even audible artifacts, going from analog to digital and vice versa so many times?

There is a test track around here somewhere that has 7x roundtrip of DAC/ADC. You can try to ABX that…

 

[Colonel7]

Subwoofers at all? Almost any 5” is going to be a compromise beyond a few feet and even at near field I prefer to cross them around 100 Hz. There are a lot of excellent linear low distortion drivers that won’t break the bank either.

What is your source? A pc, something else or do you use a DAC or preamp to switch among sources? If it’s a pc you have a lot of freedom, even do a hybrid of simple analog crossover for the pass to the tweeter/HF and Equalizer APO for a straightforward PEQ or Dsp as complicated as home brew.

This is an excellent thread by a member with a lot of experience that will get at “forward firing”

 

[BKr0n]

Why not a cheap SBC between your source and your DAC that is running a lightweight linux distro and CamillaDSP?

Was actually going to do that at one point. If I did that I would essentially just be making a music server.

Subwoofers at all? Almost any 5” is going to be a compromise beyond a few feet and even at near field I prefer to cross them around 100 Hz. There are a lot of excellent linear low distortion drivers that won’t break the bank either.

Agreed. From what @kemmler3D was saying purifi has the 8" and 10" aluminum woofers. They also have a 6.5" but out of all of them the 8 has the highest efficiency.

What is your source? A pc, something else or do you use a DAC or preamp to switch among sources? If it’s a pc you have a lot of freedom, even do a hybrid of simple analog crossover for the pass to the tweeter/HF and Equalizer APO for a straightforward PEQ or Dsp as complicated as home brew.

I use either my PC or my Fiio M17. I'm looking into at some point getting a topping dac. The idea was to source and upsample from my pc, then send it to the DAC and speakers.

 

[Colonel7]

Agreed. From what @kemmler3D was saying purifi has the 8" and 10" aluminum woofers. They also have a 6.5" but out of all of them the 8 has the highest efficiency.

Amazing linearity and distortion but you’ll be in the $3500 - $4500 per pair of DIY speakers with DSP and amplification. Going endgame right away!

 

[kemmler3D]

Say I use the topping dac as my source. Wouldn't there be loss, or even audible artifacts, going from analog to digital and vice versa so many times?

There will be losses (mainly extra noise) but very unlikely to be audible. Remember that -80dB is 0.000001% and we're actually talking about devices that are many times better than that.

EDIT: Also, why specifically 8" or 10" ? They make an aluminum that's also a 6.5" so wouldn't that be better to cut down on size?

I was being a bit silly... I also mentioned Bliesma because it has some of the lowest distortion of any tweeter I've seen. The Purifi 8" and 10" have the lowest distortion for woofers / mids that I've seen. However they all happen to be among the most expensive drivers out there. And they're big, too. That's just what you would use if you wanted minimum distortion and money is unlimited.

That said, if you go with the 6.5" or 8" you're already on a very good path in terms of distortion. Pair it with the KEF coaxials you already bought and you might have something very interesting on your hands. The Purifi 6.5" can actually go pretty low for its size, so it's considered a good starting point for a high-end bookshelf.

But realistically, you don't need Purifi drivers to get a good, low-distortion bookshelf together, they're just the pinnacle of low-distortion midwoofers these days, if you have the money.

 

[BKr0n]

That said, if you go with the 6.5" or 8" you're already on a very good path in terms of distortion. Pair it with the KEF coaxials you already bought and you might have something very interesting on your hands. The Purifi 6.5" can actually go pretty low for its size, so it's considered a good starting point for a high-end bookshelf.

I was actually eyeing up the 6.5" anyway because of the efficiency and low end extension. I think it's probably the best compromise of size to performance with good distortion figures. In it's current iteration, its the KEF with a sensitivity of 93db and the purifi with 87db. Two purifis ran at 2 ohms will make up the sensitivity difference. PRs added so I don't have to port the enclosure. From there it just the electronics (also need to redo my crossover since I think it would be a waste if I didn't at least overlap the purifis into the mid range). Much cheaper than with the RAALs . Additionally, since I would be running the purifis at 2 ohms, I can even enjoy a little overall power efficiency so no crying over my electric bill.

I guess the last question then is DSP. Out of all the thing's I've tried, sigmastudio seems to be the most flexible in what you can do at board level. A lot of companies like dayton and minidsp use it for their boards. That usually runs from one of analog's processors (sharc or sigmadsp). The adau 1701 seems to be a popular pick for most of them. There's a lot of premade solutions, kits, etc. I can build my own circuits just fine, but I also recognize the benifits of a premade solution. I guess what I'm asking is price to performance, if i wanted to eek out every little bit I can get, is it better to buy a premade, or make my own? (As an aside, I already have most of an amplifier circuit designed so really I could just use the DSP as opposed to needing one like with hypex. Although, hypex does make the DSP board all by its lonesome without the amplifier)

 

[kemmler3D]

since I would be running the purifis at 2 ohms, I can even enjoy a little overall power efficiency so no crying over my electric bill.

I think this works in terms of efficiency but a lot of amps don't work well/properly at 2 ohms, so I would be careful running them in parallel like that.

if i wanted to eek out every little bit I can get, is it better to buy a premade, or make my own

In terms of the DSP itself - I don't think there is going to be any difference. It's all digital so the quality should be basically 100% either way.

 

[BKr0n]

I think this works in terms of efficiency but a lot of amps don't work well/properly at 2 ohms, so I would be careful running them in parallel like that.

Ordinarily I would be worried about it myself, but the amp I'm making for them is going to use a TPA3255. Apparently it can run speakers at 2 ohms while remaining stable so at least I'm good on that part. At this point I'm more concerned about the power section and board construction. I've done plenty of single side boards, but no double or smd solder, but I can't imagine it's that much tougher as long as I'm careful about heat. As for power, one thing I've definitely noticed about power electronics is it's very weird in terms of price because whether you buy or make it it's still going to be expensive so gotta weigh my options there.

In terms of the DSP itself - I don't think there is going to be any difference. It's all digital so the quality should be basically 100% either way.

Theeeeeeeen I guess I'mma pick a board lol. I guess the key thing I should look for then is at least 32 bit resolution.

 

[kemmler3D]

I guess the key thing I should look for then is at least 32 bit resolution.

Internal 32/64 bit processing is pretty common, yeah I would certainly advise looking for that.