A high-end digital amplifier for everyone

[jwb]

The DAX website is offline at the moment, so I haven't looked at the drawings. Of course there will be some interest from the members of this forum. Some people may even be able to help out

I've been doing a lot of high-speed digital board layouts lately, I'd be happy to help out with those drawings and I already have invested in a multilayer CAD/CAM package, Eagle.

4-layer screened and masked boards can be had for about $40 each in quantity from the house I prefer, pcbexpress.com. Their work tends to be high quality.

It seems to me one of the important points with these digital amps is controlling the output level by modulating the suppy rail. This is also an area of interest for me and I have some ideas on the subject.

Probably the person who has written the most on the subject of digital amps on this forum is Brian Brown. He'll be interested surely.

If someone has the drawings, please forward them to me.

Cheers,
jwb

[Pjotr]

Hi Jwb,

There was no extensive explanation on the web-site so far. Hopefully they are now updating it, so there comes more in-depth information what it is all about.

PSU modulation is indeed one of the main issues and it is very hard to tackle. All PM-amps suffer from what is called “power supply pumping”. It is my vast opinion that this can only be tackled by using feedback from the output of the amp. And this is exactly what Tripath and Spectron do with their amps.

Cheers

[LineSource]

DAX: I have been pleased with the TI TAS5015 equibit chip, and hope you would consider using this chip in your PCB design.

My experimental amp uses the 240/120 AC inputs from the output stage power supply transformer to create a manually selectable 50% and 100% voltage. For detailed casual listening I switch on the 50%, for HT the 100% full voltage. The ideal design might have 2-4 voltage plateaus with hystersis. Monitoring the input digital level would allow switching up to the next plateau ahead of the loud transisent.


Digital amps are so efficient that reducing the power supply voltage is done to get more detail per bit of resolution, and not so much to reduce heat.

[DAXgroup]

you're going to have to make it easy to reprogram, no one is going to want to keep the firmware it was shipped with for long (byt the sounds of it you already wanted to do this), not that your firmware will have anything wrong with it, but everyone's going to want to change things and add other features

The board is reprogrammable using a simple home-made JTAG cable connected to the parallel port. Xilinx provides a free programming software which is fairly easy to use.

I have been pleased with the TI TAS5015 equibit chip, and hope you would consider using this chip in your PCB design.

The goal of the project is to have a platform for digital amplification development, not simply an implementation of an existing IC

It is my vast opinion that this can only be tackled by using feedback from the output of the amp. And this is exactly what Tripath and Spectron do with their amps.

Adding feedback to a true-digital amplifier is not an easy task. Since all the audio processing is done in the digital domain (fpga), feedback from the analog domain requires an ADC which adds delay in the loop. Keeping the amplifier stable becomes a very hard task. Some papers have been published on this topic, and we agree that it might be interesting to experiment with this. However, before attempting to design a feedback algorithm, it is essential to reach the limit of the open-loop topology. Let us remind that, for its feedback design, Tripath uses an analog input, it is therefore easy to add negative feedback at the input. However TI equibit amplifier's uses an open-loop topology with a digital input, this is what we consider a true-digital amplifier (that some might call a powerDAC). It is motivated by the fact that the source (CD) is digital and it is wasteful to proceed to D/A conversion in order to generate the PWM.

As for varying the power supply voltage, we also agree that it is a very good idea. As you said, it overcomes the dynamic range problems, but also significantly reduces EMI.

Within a month, we will be able to release the documention for the project, the user manuals. Maybe that will be easier at that moment to have everyone understanding every aspects of the project.

For sure there will be a lot of efforts to be put on the PSU as the current platform doesn't include a reference power supply design.

Meanwhile we want to offer our first thanks to everyone that manifested interest.
best regards,
DAX Group

[Petter]

Originally posted by DAXgroup


The goal of the project is to have a platform for digital amplification development, not simply an implementation of an existing IC

[SNIP]

Adding feedback to a true-digital amplifier is not an easy task. Since all the audio processing is done in the digital domain (fpga), feedback from the analog domain requires an ADC which adds delay in the loop. Keeping the amplifier stable becomes a very hard task. Some papers have been published on this topic, and we agree that it might be interesting to experiment with this. However, before attempting to design a feedback algorithm, it is essential to reach the limit of the open-loop topology. Let us remind that, for its feedback design, Tripath uses an analog input, it is therefore easy to add negative feedback at the input. However TI equibit amplifier's uses an open-loop topology with a digital input, this is what we consider a true-digital amplifier (that some might call a powerDAC). It is motivated by the fact that the source (CD) is digital and it is wasteful to proceed to D/A conversion in order to generate the PWM.

As for varying the power supply voltage, we also agree that it is a very good idea. As you said, it overcomes the dynamic range problems, but also significantly reduces EMI.

DAX Group

I looked into the work by Sandler et al several years ago and came to the same conclusion TacT seemed to capitalize on first. I feel your comments above precisely illustrate the salient issues.

At the same time, it would be interesting to benchmark against say the TAS5015 for comparison purposes. The problem with the TAS unit is general availability as well as no internal changes possibly by the DIY crowd. Your solution to this is very cool!

I also wonder if it is possible to use your board as a DAC unit unit - certainly your modulator is interesting, and if you could modulate say 64 outputs swithing at MHz, adding some randomness and having say 4 levels - all tweaked analog current sources, my belief is that you could make a killer DAC unit.

Petter

[DAXgroup]

Making a DAC out of the current platform is fairly easy. However, the output stage would have to be changed, but since we designed the platform in a modular fashion, this is a simple task.

DAX Group

[Petter]

Consider attached drawing.

I have pondered over using an FPGA with a lot of outputs, some randomness (as seen by a dCs) + a couple of levels (say 4-24 units per bank, 2-4 levels in total ). I never had the courage to start learning FPGA programming - but only having to modify a project is another kettle of fish altogether

I would expect switching at MHz interesting

Output stage is for demonstration only. JFET's and matched current sources are probably cheaper than ultra-precision resistors.

Let me know what you think

Petter

[Jaka Racman]

Hi,

I am very interested in the design, but I would like to make a suggestion. With a platform like this it would be very easy to make multiphase output stage. Advantages of such approach can be seen in the following US patent applications:

20020053945 (Bruno Putzeys, Philips) and

20030095013 (John Melanson, Cirrus Logic)

So I suggest a minor modification to the modulator board:

Expand output stage I/O header from 20 to 40 pins (I suppose every other pin on header is GND.) Therefore it would be possible to drive more phase shifted output stages in paralell with lower per phase frequency and corresponding lower distortion due to timing errors. It would also make possible to use 80 pin parallel ATA (hard disk) cable with extra 40 ground wires and greater timing fidelity to connect to the output stage.

Best regards, Jaka Racman

[DAXgroup]

With a platform like this it would be very easy to make multiphase output stage. Advantages of such approach can be seen in the following US patent applications:

Hi, this is very interesting. Do you where we could find information/papers on those patents?

On the hardware side, there are still a lot of pins available on the connector between the fpga board and the output stage. It would be easy to design a different output stage board with two TAS5112 chips and run them with different PDM signals. You could also use two seperate output stage and hook them to different header connector on the fpga board.

Regards,

David
DAX Group

[Jaka Racman]

Hi,

patent applications can be accessed at US Patent Ofice http://appft1.uspto.gov/netahtml/PTO/search-bool.html .
Just type in patent application number. To access images you will also need tiff viewer http://www.uspto.gov/patft/help/images.htm .

Regarding proposition for 40pin header it was more in direction of signal fidelity than in the number of the I/O pins, since I realised you have 79 I/O pins on 4 20 pin headers. I have some expirience with analog multiphase class D amplifier that had modulator and power stage connected by 40cm ribbon cable. Although every second wire was GND, I had some noise from power stage coupled back to modulator board. Result of this jitter was relatively noisy amplifier. Multiphase designs are particularly susceptible to noise because you have two stages switching at the same time with zero input signal when using even number of phases. This is why I use LVDS transmitters and receivers together with CAT5 cable in my next amplifier design (still in very early stage). I thought of using 3 TAS5015 modulators each fed with its own SRC1942 from the common input signal. I would then phase shift LR clock of each TAS5015 and then hopefully have phase shifted PWM signals. But I would give all that up in favour of DAX.

Best regards, Jaka Racman