ASR Open Source Streamer Project

[tinnitus]

Khadas Tone Board is only 2 channel stuff

 

[somebodyelse]

In my head, I see this as something that could be cobbled together out of four linked V1 KTBs, a Pi Zero, and DSP software in an open source 3D printed enclosure.... that's 8 channels out with impressive SINAD for a couple hundred bucks.

How do you intend to link the Tone Boards? Custom firmware could do it, but it's a step up in complexity from most of the proposals so far, and I'm not sure the XMOS licensing would allow it to be open source. Or you need a USB to 4xSPDIF adapter - are there any inexpensive ones? The Pi Zero is a bit lacking in horsepower to do much DSP too, unless you use a hat such as the DAC+DSP. It's not entirely clear whether it suffers from issue #2215 so you might have pops and clicks to contend with if you're using USB audio. And it's going to be a bit more than a couple hundred bucks unless you've got a cheap source of KTBs.

 

[notabenem]

What is the general idea about getting the sound of movies to the speakers? Are you considering HDMI input/output (and some basic or fancy multichannel formats - DTS MA etc)? It would be great if this could replace my (old) AV receiver.

 

[dwkdnvr]

I actually spent a bit of time looking into the current state of RPi (well, Linux/alsa) audio options. I came away disappointed/frustrated for the most part. A truly state-of-the-art multi-channel DSP system based around the RPi will likely require some custom hardware, or else just throwing in the towel and using the Okto DAC8 Pro.

As @sombodyelse points out, managing output to multiple cheap DACs is more work than it first appears. There's the MiniDSP UDIO-8 which gives 4xspdif in/out, but it was a bit underwhelming in Amir's test - rather high jitter. Plus, it's $325, and a NanoDigi is only ~$200, so if you don't need inputs you might be better off just going with a simple streamer with stereo out and implementing the DSP in the NanoDigi. Even so though, you're looking at 600-750 by the time you add 4xKTB which puts you up into the price range of the pro audio interfaces which give you even more flexibility with potentially comparable performance in a single box, and you're only a couple hundred shy of the Okto which would seem to be a no-brainer in comparison.

Diyinhk has an ESS9038 based 8-channel DAC kit using the Xmos interface which should be class compliant and work fine under Linux. https://www.diyinhk.com/shop/audio-...hz-32bit-es9038pro-pcm-dxd-dsd-audio-dac.html This kit is only $200 with xmos interface, and so might be the leading candidate if a one-box solution with custom enclosure is a high priority. Plus, it's 'likely' that a bit of hardware work would allow interfacing one or more A/D converters to the XMOS board to provide some input functionality.

The big disappointment I discovered though is that even though the Motu interfaces are "class compliant" it seems they don't really work under Linux at this point - they "work", but quickly run into channel mapping and distortion problems. (this is the AVB / 8A / 8D range - it looks like the M2 and M4 work fine with the latest kernels). Focusrite Scarlett interfaces and the Behringer UCM units all appear to work fine, but they're a step down (or two steps in the case of Behringer) in performance from Motu. Using the MiniDSP USBStreamer to get ADAT in/out and pairing that with a Motu 8a or 8d looks like it would be workable, but it wouldn't satisfy the one box/ custom enclosure requirement.


TL/DR So, after all that it seems that I'm rather uncertain on where I'd suggest going with this idea. Options
- 4 channels with input capability the Motu M4 seems to be the best fit. Would basically allow re-creating a MiniDSP SHD aside from the lack of digital inputs and would run ~$300 including an RPi 4. I actually have all this in-house, and so may actually start tinkering.
- 8 channels 'high end' would be to use the Okto DAC 8 Pro. It offers AES/EBU inputs, but they're problematic since they have to be clock masters, and so this would make most sense if you didn't really need inputs. Truly state of the art, but a 2-box solution. figure $1300 or so
- midrange or better >4 channels with inputs - a pro audio interface. The Motu AVB interfaces are what I'd want to go with, but that would need diving in to address the driver issue with unknown prognosis - some capable people have already been looking at it without success. The Focusrite Scarlett interfaces just don't excite me but they may be the best option for 'out of the box' if you need more than 4 channels - $500-700 and again a 2 box solution
- the DIYINHK ESS 9038Pro kit is the best thing I've found so far for a more or less ground-up custom approach. This obviously assumes that their implementation is at least competent, and requires assembling the kit which may leave some people out. Base parts cost is pretty reasonable though - maybe $400 plus the enclosure.

 

[dwkdnvr]

What is the general idea about getting the sound of movies to the speakers? Are you considering HDMI input/output (and some basic or fancy multichannel formats - DTS MA etc)? It would be great if this could replace my (old) AV receiver.

HDMI input is pretty much a non-starter for DIY / open source projects. Too much DRM involved. My read on what is being discussed is that this is a 2-channel music focused device. The only or at least most obvious way this might participate in a multi-channel system would be to have a special 'HT Bypass' mode where you feed in the main L/R signal from the AVR (and maybe the LFE as well) and have it pass through the DSP crossover to drive the mains (and possibly subs, if we're pushing the functionality that far). It won't replace an AVR since it won't be able to decode any of the multi-channel formats found in modern content.

 

[buz]

Well put, pretty much matches my conclusion.

Final option would be to try and convince one of the companies making well measuring affordable DACs to engineer a reasonably priced 8 channel USB board. Thinking in particular of khadas who has experience with maker products (and actually makes SBCs...) but maybe also Topping.

Or a cost down Okto8 without bells and whistles. One can dream

 

[notabenem]

If we get into the $700+ range, then it's getting dangerously close to the Denon X3700H with its (additional) HDMI and 11 multichannel preamp capabilities. While its DAC is measurably sub-par compared to the mentioned stand-alone DACs, its believed to be likely inaudible once it reaches the speakers (based on https://www.audiosciencereview.com/...active-monitors-dac-and-dsp.16435/post-690601). At this point the business case is getting rather awkward, isn't it?

 

[dwkdnvr]

If we get into the $700+ range, then it's getting dangerously close to the Denon X3700H with its (additional) HDMI and 11 multichannel preamp capabilities. While its DAC is measurably sub-par compared to the mentioned stand-alone DACs, its believed to be likely inaudible once it reaches the speakers (based on https://www.audiosciencereview.com/...active-monitors-dac-and-dsp.16435/post-690601). At this point the business case is getting rather awkward, isn't it?

Yeah, HDMI out to an AVR is definitely an option as well - doesn't fit Amir's originally stated requirements, but the same software stack could be used this way. Certainly would have the cheapest overall system cost.

I know the RPi originally was limited to 2-channel output over HDMI, but I believe this has been fixed at this point.

 

[notabenem]

HDMI is pretty much a non-starter for DIY / open source projects. Too much DRM involved.

If the same RPi box would be able to act also as a (video) media-player (e.g. VLC, Kodi) then the decoding of most of the formats is given (except some fancy object based ones) and HDMI input is not needed. At that point the only the DAC channel count is going to affect the price in a significant way.

 

[MatthewS]

I think the issue is that those MVPs already exist in several forms.

Can you point me to one that is affordable and measures well?

RPi likely means skipping Chromecast and Airplay2 unless we/someone tackles manufacturing a hat with the required hardware and licensing.

https://github.com/mikebrady/shairport-sync/issues/535

 

[dwkdnvr]

If the same RPi box would be able to act also as a (video) media-player (e.g. VLC, Kodi) then the decoding of most of the formats is given (except some fancy object based ones) and HDMI input is not needed. At that point the only the DAC channel count is going to affect the price in a significant way.

Well, the impetus for this was specifically for a DSP crossover to accompany the Directiva speaker project. Kodi/VLC etc can bitstream any format down to the AVR over HDMI, but if we're trying to decode it to PCM in order to apply our crossover DSP, then it's a problem - there is no way that I know of to decode any of the modern DTS/Dolby formats in open source software. Maybe I'm overlooking something, though.

 

[notabenem]

Maybe I'm overlooking something, though.

quoting from https://forum.kodi.tv/showthread.php?tid=302674&pid=2490398#pid2490398 :
We have lossless decoders of DTS-HD, TrueHD, AC3, DTS and many more. The only thing we currently cannot decode is the Atmos bits.

 

[notabenem]

Well, the impetus for this was specifically for a DSP crossover to accompany the Directiva speaker project.

You made me think even more. My understanding is that those speakers will have the DSP in them (and thus a DAC too). In this case maybe we don't need a DAC in the streamer at all (maybe only as a 2 channel fallback). Much more a set of digital outputs to the speakers and thus save a couple of DA->AD round-trips.

 

[Raindog123]

...Basically just using the pi as a monster FPGA for DSP

Can you please elaborate on how a Pi - without any FPGA chip - would become a “monster FPGA”? Through a custom hat?

More generally, can we size the processing requirements for this project? (Are Pi’s BCMs powerful enough for our needs? Do we actually need an FPGA for DSP?)

In my line of work we commonly use Zynq SoCs (eg Zynq 7000 and Ultrascale+) that combine both a general purpose processor and FPGA fabric in one chip. There are multiple development and embedded production boards (eg, Zedboard, iVeia products, etc). But those might be an overkill here, both performance and price wise.

 

[buz]

You made me think even more. My understanding is that those speakers will have the DSP in them (and thus a DAC too). In this case maybe we don't need a DAC in the streamer at all (maybe only as a 2 channel fallback). Much more a set of digital outputs to the speakers and thus save a couple of DA->AD round-trips.

The DSP,vDAC and AMP will be outside the speaker so this would be the DSP, DAC part. The other option for that presumably would be a miniDSP.

 

[frangle]

The DSP,vDAC and AMP will be outside the speaker so this would be the DSP, DAC part. The other option for that presumably would be a miniDSP.

We seem to be drifting from the original idea:

4. Support for active speakers using software crossovers and multiple external DACs.
8. Availability of a complete solution in a case ready to go (sans DACs of course).

 

[buz]

We seem to be drifting from the original idea:

Software cross over is why we need the multiple DAC channels in the first place...

 

[notabenem]

Apparently certain components, their roles/responsibilities and placement are unclear.
Have a look on this diagram, as a fodder for thought (and feel free to change, edit as you see fit)
https://mm.tt/1809494229?t=aVmjf6osnx

My impression is that yellow bubbles need to be clarified - the link between this streamer and the speakers (Directiva or not).

 

[somebodyelse]

The DSP,vDAC and AMP will be outside the speaker so this would be the DSP, DAC part. The other option for that presumably would be a miniDSP.

Or a couple of FusionAmps.

 

[buz]

Or a couple of FusionAmps.

That would work, too. Would need to provide a case for them as there is no space in the current directiva prototype