New low-cost DSP platform in development

[bachatero]

At this point, the DSP functionality is an irrelevant detail because all that's needed is to implement the DSP math which likely isn't a huge challenge once the foundational language part is done. The reason why this is a bigger deal on other platforms is because you have to take it or leave it with regards to their effect selection, but that's not the case here since you can use the standard library's effects or your own. The most important aspect of this project aside from that is the supporting cast, IO and UX and price. The competition also falls flat here and the core DSP functionality depends on it so that's why I'm heavily focusing on that right now.

 

[bachatero]

Time for a Christmas present! To hone my hardware engineering skills for this DSP device, I want to first make a very simple and low cost DAC using top of the line hardware. But how?

It turns out that in "bulk" quantities of just 25, the very best ESS DAC chips are $14 each. Combine that with a USB to I2S/SPDIF chip, microcontroller, opamps, ports, PCB, and other cheap parts, I think we might be able to squeak by with 120dB SINAD for less than $80. That's because I have zero (0) employees and electronics R&D has gotten very cheap, like really cheap. Thoughts?

 

[voodooless]

It turns out that in "bulk" quantities of just 25, the very best ESS DAC chips are $14 each.

They are not:

https://octopart.com/es9039spro-ess-138330195

Sure, the lower-end models are fine and much cheaper, but the “very best” is not $14 each…

Combine that with a USB to I2S/SPDIF chip, microcontroller, opamps, ports, PCB, and other cheap parts, I think we might be able to squeak by with 120dB SINAD for less than $80. That's because I have zero (0) employees and electronics R&D has gotten very cheap, like really cheap. Thoughts?

Implementation is everything. You’ll need the knowledge and experience to make this work with few iterations. For way less than $80, you can get ES9018k2m boards, in fact, here is one for € 12 The chip can also do a 120 SINAD. The only issue is finding a board with a decent implementation, but at these prices you can buy plenty of boards to test for the same money as 25 of those $14 chips. Now, as a fun learning experience, sure, knock yourself out! But given the software challenges, it seems you already have plenty on your todo list.

 

[bachatero]

Sure, the lower-end models are fine and much cheaper, but the “very best” is not $14 each…

I just bought 5 of these at $14.67 each though:

https://www.mouser.com/ProductDetail/460-ES9039Q2M

The 8 channel one you linked can only achieve their maximum performance (which is just 2dB better) when combined in mono mode, so that would be $70 * 2 = $140 just for the chips. That's a total nonstarter!

 

[voodooless]

I just bought 5 of these at $14.67 each though:

https://www.mouser.com/ProductDetail/460-ES9039Q2M

The 8 channel one you linked can only achieve their maximum performance (which is just 2dB better) when combined in mono mode, so that would be $70 * 2 = $140 just for the chips. That's a total nonstarter!

The “very best”, means top of the line. ES9039 pro is that. It obviously is a non starter, and I would not recommend them. The Q2M version you chose is perfectly fine obviously. It’s just not “the very best”. Not that you’d hear the difference though

 

[MCH]

Time for a Christmas present! To hone my hardware engineering skills for this DSP device, I want to first make a very simple and low cost DAC using top of the line hardware. But how?

It turns out that in "bulk" quantities of just 25, the very best ESS DAC chips are $14 each. Combine that with a USB to I2S/SPDIF chip, microcontroller, opamps, ports, PCB, and other cheap parts, I think we might be able to squeak by with 120dB SINAD for less than $80. That's because I have zero (0) employees and electronics R&D has gotten very cheap, like really cheap. Thoughts?

My thought is that your final budget is going to get out of hand. I thought your objective was something cheaper than the entry level mindsp. I think that will be pretty impossible with 14$ ICs... My advice: check which DAC chips minidsp uses in its last, much more expensive, multichannel devices. They are less than half the price of what you ordered. My 0.02c

 

[olieb]

It turns out that in "bulk" quantities of just 25, the very best ESS DAC chips are $14 each. Combine that with a USB to I2S/SPDIF chip, microcontroller, opamps, ports, PCB, and other cheap parts, I think we might be able to squeak by with 120dB SINAD for less than $80.

That is already there. Well "only" 116dB (not that it mattered), but finished product with case, warranty and all that.
Didn't you say something about wanting cost going down?

SMSL SU-1 Stereo DAC Review

This is a review and detailed measurements of the SMSL SU-1 stereo USB DAC. It was arranged to be sent to me by shenzhenaudio from the company and costs US $79.99. Despite its simplicity, the SU-1 manages to look elegant with nice fonts and LED indicators for different inputs. Back panel has...

www.audiosciencereview.com

I am just so relieved that the splash screen design ideas for the web page have already been presented. Otherwise I might even think this project is just a balloon of hot air.

 

[KSTR]

Time for a Christmas present! To hone my hardware engineering skills for this DSP device, I want to first make a very simple and low cost DAC using top of the line hardware. But how?

Be warned, TOTL hardware needs TOTL skills to achieve TOTL performance, including TOTL measurement gear. That is especially true for ESS chips because of their abundance of setup registers/parameters and various operating modes.
Clocking is everything as is proper supply and reference voltages, and not to forget the layout. You can get away with a single LT3045 (as those aren't cheap either) for 3.3V supplies and references if you really know what you are doing (note: better have network and spectrum analyzers at hand). At least the I/V-stages aren't a mystery anymore since ESS included usable schematics in the datasheets.

 

[jbags]

Be warned, TOTL hardware needs TOTL skills to achieve TOTL performance, including TOTL measurement gear. That is especially true for ESS chips because of their abundance of setup registers/parameters and various operating modes.
Clocking is everything as is proper supply and reference voltages, and not to forget the layout. You can get away with a single LT3045 (as those aren't cheap either) for 3.3V supplies and references if you really know what you are doing (note: better have network and spectrum analyzers at hand). At least the I/V-stages aren't a mystery anymore since ESS included usable schematics in the datasheets.

 

[jbags]

Bachatero, you are very energetic and keep up the good work. Just don’t forget to charge something for all your effort. Best wishes.

 

[bachatero]

The reflow oven has arrived, now the ESS and TI chips are ready for sampling!

 

[KSTR]

 

[bachatero]

View attachment 417586

Almost completely a nothing burger. Plenty of PCBs already have a solid ground plane, and the rest is common sense for signal integrity.

 

[LionIT]

The good intentions of this project seem blurred to me, but maybe I am missing something...
Has the milestone of a cheap versatile DSP with reliable digital interface been reached?

 

[bachatero]

The good intentions of this project seem blurred to me, but maybe I am missing something...
Has the milestone of a cheap versatile DSP with reliable digital interface been reached?

I am working on the software side as we speak but it turns out there is no good foundation for the code generation because nobody's done this before, so I have to "reinvent the universe" first. But at the same time, on the hardware side because although this has been done before, the previous work isn't open source so I have to "reinvent the universe" there too.

 

[LionIT]

I am working on the software side as we speak but it turns out there is no good foundation for the code generation because nobody's done this before, so I have to "reinvent the universe" first. But at the same time, on the hardware side because although this has been done before, the previous work isn't open source so I have to "reinvent the universe" there too.

Ok, thanks.
Before reinventing the universe shouldn't you make sure you can get reliable hardware?
I mean, a DSP with digital interface is the basis, but with it there are countless problems to solve (Clock, buffer, noise, emi, power supply).
Taking Chinese boards online is a lottery sometimes, and it is not enough to see that they "work". I tell you from experience.
Some time ago I bought an I2S to SPDIF / Toslink interface. It worked somehow. But when I connected it to my audio interface with Toslink there was a terrible noisefloor measured with RTA, and dependent on that board's location. If I approached it to metal components, the noise floor worsen and vice versa. And this was all in the digital domain!
In the end I realized that the board was hypersensitive to the clock i2s cable, and there was no way to make it immune because it was badly designed (not even with those micro coaxial cables).
So, I bought just the same board that you are using and I no longer encounter such problems. However, with this new board after a while my interface stopped locking on the Toslink clock. I didn't have the necessary oscilloscope to see what was happening, but it was probably a limit of rise time or something like that...
So I tried another one then, but the voltage regulator of the mother board from which I2S signal was sourced and that powered the I2S / Toslink board was not stable (resonance) due to some component of that board and created noise on the clock, manifested as a very high jitter.
From further investigation it seemed that the problem was some ferrite filters matched with ceramic capacitors.

Well, in short ... I trashed everything and bought professional hardware in the end.
Too bad I wasted a lot of time (not many money) trying to run electronics designed in a not entirely serious way and without know requirements/effects of the rest of the system.

This is not to invite you to let it go, on the contrary I really like the idea, but to suggest a more precautionary approach for your precious time!

PS. Out of curiosity, have you tried to see what functionality you can achieve with the new DAC ES9082 with integrated DSP and line driver?

 

[KSTR]

Almost completely a nothing burger. Plenty of PCBs already have a solid ground plane, and the rest is common sense for signal integrity.

Well, I can understand why you are using these adapter boards which violate all this common sense -- to get the stone rolling -- but OTOH I feel this is a somewhat wasted effort as you won't land anywhere near tre the performance the ES9093q2m can deliver... but maybe that isn't your goal after all?

 

[Bamboszek]

I feel like we're seeing a perfect example of the Dunning-Kruger effect here.


Don't get offended, but it looks like it's the very first PCB you designed with basic rookie errors like vias covered by silkscreen and soldermask.
This is far, far away from skill required to achieve performance numbers quoted in the datasheet.

Designing high performance mixed signal circuits such as audio DAC is a huge and hard task on its own.
Proper DAC design would probably take year or so, and this is not even main goal of the project.

Instead of designing hardware, optimizing compiler and so on, I would suggest to take more iterative approach.
Start with basic HW and SW platform which is already available and proven to work. Prepare some proof of concept of any DSP operation with and two channel SPDIF or whatever would be easiest to implement. Once you have a working example, you could add anything you think is important step by step.

Otherwise, in my opinion you would end burned out in development hell without any practical outcomes.

 

[bachatero]

@Bamboszek Talk is cheap, are you the one writing assembly code and reflowing solder? I already fixed the mask issue thanks to custom 3D printed solder paste stencils made in OpenSCAD.

 

[cobby]

it would be cool if you could explain your concept a bit more.
would like to know whats the benefit of permanently compiling each filter, if thats the idea? sounds like a good idea somehow but i havent got the concept and how it compares to exisiting solutions.

for hardware i would really stick with exisitng stuff for now, like it was sugested here.