Advice request - evaluation board with low latency and high processing power

I’m looking for an evaluation board to develop an acoustic active noise cancellation for fan noise in different conditions (duct with long or short propagation propagation path). I have some experience in this field and previously developed FXLMS on an ARM board and on an old Texas Instrument DSP but these implementations had some limitations especially due to latency and lack of processing power.

In this project, I would like to test some:

  • Ultra-low latency implementation such as in the headset (simple feedforward or feedback algorithms), which seems to be achievable with some of your ADACs.
  • Hybrid algorithms such as the FXLMS, eventually MIMO on small propagation distances which will require much more processing power and lower latency than what I currently have on-board.

So, an evaluation board with a top-notch DSP seems the way to go but there are so many references  that a little help will be appreciated.

I’m looking for a solution that could manage ultra low latency (such your ADA converters with DSP for headsets) and high processing capabilities.

I’ve narrowed the choice to three boards and I would like which one would be the best choice or if you have any better suggestion.

These are the :

  • SHARC audio module,
  • ADZS-SC589-EZLITE
  • ADZS-2159-EZKIT.

 Is this an adequate selection and which board would be best suited for my needs?

For what I understand:

  • The SHARC Audio module has the converter ADAU1761 and has one stereo input and output (no MIMO control then, except if it can be easily extended) but has the Metal Bare Framework for easier programming.
  • The other boards have more input/output and the converters ADAU1979 for Analog to Digital and ADAU1962A for Digital to Analog. But the Bare Metal framework is only available for the Audio module, am I right?

 

Thank you in advance for your help.

  • 0
    •  Analog Employees 
    on Oct 20, 2020 6:10 PM 1 month ago

    Hello SLesoinne,

    If you need really low latency and a device that is designed for this application then I would give a series look at the ADAU1787. It has a fast core that can run at 768Khz fs, with built in ADCs and DACs for an analog to analog latency in the single digit microseconds. I think it is around 5 or 6us. 

    Then it has a DSP core that allows for the monitoring of the process and to perform some of the slower calculations to then change the coefficients in the fast DSP. It has bank switching for easy switching between filter coefficients. 

    Any part that has external ADCs and DACs will have the serial port latency for transferring the data from the ADC to the DSP and then from the DSP to the DACs. Then you have the latency of the DSP. So that cripples the performance before you have written a single line of code. 

    The ADAU1787 is also VERY small and very low power. 

    The only issue is that we cannot help you very much with the actual algorithm since we work with many customers and do not want to step on any toes. 

    Dave T

  • Hello DaveThib, thank you for the fast answer.

    I totally understand the algorithm and IP-related "issue".

    By the way, latency around 5 or 6 µs is really really fast (propagation distance of 2 mm) and could lead to some really interesting design for my tests.

    I'm also planning to test other configurations where the latency doesn't need to be so extreme. Do you know what order of magnitude I can expect for the latency when using the other aforementioned boards (or any other with communication between the ADAC and the DSP).

    Just a little more details for clarity: The field of testings would cover HVAC systems that can range from centralized (propagation delays are not really a limitation) to decentralized (propagation delays really do matter). One point of attention with the ADA1787 is the use of IIR filters (64 if I'm correct) and I have some doubt about it's ability to achieve sufficient compensation of the back-propagation path from control source to the reference microphone (feedforward algorithm) which can degrade the performance on broadband signal control.

    Stéphane L.

  • 0
    •  Analog Employees 
    on Oct 21, 2020 2:05 PM 1 month ago in reply to SLesoinne

    Hello Stephane,

    First I need to mention that I do not support the SHARC platform so I am not the best one to comment on those processors. I support the SigmsDSP products. If the PCB layout is done well then you can run the serial interface at a sampling rate of 192kHz, then it takes 5.2 us for the data to shift out of the device. So if you are going from an ADC to a DSP then to a DAC you will have an overhead of 10.4us just with the data transfer. Then in the SigmaDSP cores you will have some internal delays of three more sample periods to get to the core, through the program and out to the serial ports. So add another  15.6us for a grand total of 26 us. 

    Then you have to look at the group delays of the ADC and the DAC filters. For our low latency codec, the ADAU1372, the group delay at 192kHz for the ADC is around 60us. But, this does include the serial port delay to shift out the data. For the DACs it is around 80us. So if you use a SigmaDSP you are around ~156us analog to analog. I could have missed a detail or two here but it gives you an idea. This is without any filtering of the audio in the core, that will of course give you some more latency. 

    That is interesting about the centralized systems. I had never realized that you can benefit from the time it takes for the plane waves in the ductwork to travel is an advantage for noise cancellation. With this in mind, you should also investigate one of our newer technologies of A2B. You can have many mics and many DACs spread out over a distance all connected by a wire in a bi-directional network. This would be very good for things like centralized HVAC systems. I could hook you up with someone to help you with that technology if you are interested. Here is a short video on the topic. https://www.analog.com/en/education/education-library/videos/4476406387001.html

    Dave T

  • Hello Dave, thank you for this interesting answer with a really good insight of the minimal delays to expect with the ADC, DAC and transfers.

    First of all, I want to avoid any hardware implementation  at the moment due to time constraints on the project but I would like to have some demonstrators covering various user cases and ANC technologies.

    The order of magnitude around 156 µs  (+- 5 cm) (+ filters delay) is more than reasonable the tests I have in mind. I guess that my low-latency definition was less strict than yours but the solution based on ADA1787 seems interesting to me for a very compact solution in very small enclosure (thus, its use in headset) where source, control source and ref or error signals are really close together (limited de-phasing due to propagation, at least at low frequencies). This could definitely be of practical use and very interesting to know the existence of such a low latency solution.

    I previously have worked with FIR ANC for stability and phase linearity reasons but, then I should have a closer look on IIR-based algorithms to clear my mind. And concerning the back-propagation path compensation, I suspect that it could be solved with another strategy.

    Concerning the A2B, I will keep the suggestion for later as I'm in the early phase of the project but I will definitely watch the video and come back to you if needed.

    I think I should start with a more versatile board at first and switch to a compact version (ADA1787-based) a little bit later in the project. I would really appreciate if you could you hook me up with someone to help me with the SHARC board.

    Stéphane L.

  • 0
    •  Analog Employees 
    on Oct 21, 2020 6:14 PM 1 month ago in reply to SLesoinne

    Hello Stephane,

    I would look into the ADAU1452 eval board to get used to SigmaStudio and the SigmaDSP platform. The other one you should look into is the ADAU1761. This DSP core is the same SigmaDSP core in the ADAU1787 but there is no fast core. So it is not exactly the same but there are built in converters. So if you are looking for a more powerful processor that can do a lot of FFT calculations then definitely look into the ADAU1452. 

    I will mention your post to someone who could point you in the right direction for the SHARC products. I know that the SHARC Audio Module is a great product with a really low entry price for the evaluation board. Plus, it is design with audio in mind. 

    Thanks,

    Dave T