WM8805 support

[Børge Strand-Bergesen]

Hi guys,

I've just pushed my WM8805 code to Github. See
https://github.com/borgestrand/sdr-widget/blob/audio-widget-experimental/src/wm8805.c

The new codebase works both for the existing hardware (AB-1.0, AB-1.1,
AB-1.2, Henry Audio USB DAC 128, ..mkII) and for my prototype extended
with the WM8805. The schematic is a bit messy, I hope to have it too
out soon. A handfull of GPIO has changed.

There are quite a few modifications. The most important one for
non-WM8805 use is probably the addition of global variable
input_select. And for the new HW I've also added a sample rate and
input indicator using three RGB LEDs on the front panel.

Feel free to build the code and test it on the hardware you have. Read
Makefile before you do so. It defines which hardware is being
targeted. The code is functional on both hardware versions and with
UAC1/2. I'll be doing more long-term testing in the coming weeks.

The only thing I haven't tested is the Linux Quirk and its automatic
fix. That functionality has probably changed now. Is this still an
issue with modern kernels?

Cheers,
Børge



Here's the somewhat longer message which I plan to publish here and there:


Using the WM8805 as an SPDIF receiver in software control mode

Here is an engineer's viewpoint and experience on programming the
WM8805. I've decided to share how I got this chip to work well on all
relevant audio sample rates (including 176.4 and 192ksps). Hopefully,
this will be a good complement to the datasheet. You may ask why I put
this together. Basically, it was harder than the toughest suduko I've
ever solved, and it felt much more meaningful. I may put the
technology into a product one day in the future, but until then it's a
very nice addition to my design toolbox.

The code can be found at:
https://github.com/borgestrand/sdr-widget/blob/audio-widget-experimental/src/wm8805.c

The WM8805 from Wolfson / Cirrus is a multi-port SPDIF receiver and
SPDIF transmitter for consumer mode signaling. (I.e. no AES/EBU.) Its
main feature is a digital fractional PLL for clock recovery. Most
other SPDIF receivers use an analog PLL structure which depends on
external passive components. Proper listening tests will come next.

This work on programming the WM8805 was is add-on to the Audio Widget
project. Just like the Audio Widget project, the commercially
available Henry Audio USB DAC 128 mkII and the SDR-Widget project,
source code is open source.

The WM8805 and the WM8804 are difficult to use. But the good news is
that the code base here makes the WM8805 work very well with two
different SPDIF sources and all sample rates. I have described some
shortcomings of the chip, but to a great extent they can be overcome.
The code handles automatic detection and re-configuration of the
notorious 176.4ksps sample rate along with 44.1, 48, 88.2, 96 and
192ksps. Another piece of good news is that the I2C / two-wire control
worked right out of the box with the routines in the AVR32 code base.
That's a first for me. I'm a bit too used to bit-banging poorly
implemented I2C....

The hardware has been tested with SPDIF and TOSLINK inputs up to
192ksps. The hardware is essentially a modified Henry Audio USB DAC
128 mkII design where a few IO lines have been reassigned. For an
experienced electrical engineer, detailed knowledge of the hardware
platform is probably not needed in order to port this design.

The devil is in the details. The sample rate detector of the WM8805
uses the same status information to encode 44.1 and 48ksps, the same
for 88.2 and 96, and the same for 176.4 and 192. In itself this isn't
that bad, if it wasn't for the fact that 192ksps needs a different
initial PLL setting from the other five. So there's basically no way
to determine from the chip itself whether that setting is needed. The
sample rate detector inside the WM8805 can't be trusted, and neither
can its sample rate change interrupt.

Luckily, the WM8805 sends out an I2S word clock of the correct sample
rate, even though the PLL isn't correctly set up. This means the
external sample rate detector can be used to determine which PLL
setting to apply. The WM8805 interrupt is polled in order to determine
whether it has lost lock.

The pivotal part of the code is the external sample rate detector. It
was programmed in AVR32 assembly code. (Yes, I cheated and looked at
the .asm which some sample .c compiled into. With that as a starting
point and a lot of time having coded other assembly languages it was
easy hacking even without knowledge of this particular machine
language.) The delays were verified with a square wave generator and
later with actual I2S. Change the MCU frequency and all the delays
must be redone.

The FreeRTOS scheduler of the Audio Widget project is halted for a bit
more than one audio word clock period, during which time a counter is
decremented. If you wish to port this effort to a different platform,
the first thing to do is determine if have the resources available to
implement a good sample rate detector. Then you must understand the
function wm8805_srd(). The .c -> .asm recipe is in its comments.

Because the starting point for the design is an USB DAC, there is code
in place to make the WM8805 and the MCU based USB audio receiver share
one I2S interface to the DAC. The RTOS software uses a semaphore to
determine which task (WM8805 monitor or USB audio) may control the I2S
and MCLK multiplexers. If your goal is to only use the WM8805 part of
the design, the semaphore code can be easily designed out.

Another shortcoming of the WM8805 is the lack of a user controlled
mute function. I'd like it to generate some functional and muted I2S
when it's trying to obtain lock or doing other things without valid
digital audio arriving. There is no soft mute or anti-pop
functionality which one can pretty much expect in modern digital
audio. Instead the code switches in the USB receiver's muted I2S
output whenever a WM8805 mute is needed. The USB receiver's output
buffer is often cleared in the USB code. That means there are few
explicit calls to wm8805_mute(). And those that are there can probably
be removed.

The current implementation of wm8805_mute() is crude. At least it
should wait until a sample boundary before setting the I2S data line
to 0. That is because a small negative value truncated with zeros will
become a large negative value to the DAC. With no control of the exact
time when the USB's I2S output replaces that of the WM8805, this
condition is hard to prevent, and some pops may be expected. A modest
hardware fix would be to have an AND gate zero the I2S data line, and
have its control signal be latched by a negative transition of the
word clock. A fuller software fix would be to route the WM8805's I2S
output into the MCU's ADC interface so that gentle muting and I2S
multiplexing can be done in software. My prototype hardware is
prepared for this option but I haven't yet begun building it.

The WM8805 in this design supports two SPDIF inputs (one over a
TOSLINK plug). When USB audio is playing the WM8805 is shut down. But
when that is not the case, the WM8805 scans its inputs until it finds
one with a lock. With lock it sends that input to the DAC's I2S input.
The WM8805 spends 200ms or so trying to lock to one input before it
gives up and tries the next. This can easily be extended to more than
two SPDIF inputs into the WM8805. This automatic input selector is
working well. The only flipside is that it doesn't respond immediately
to digital audio appearing on an input. If that is desired, the
number of inputs must be reduced, or the user must press a button for
channel selection.

While the scan mode is quick, the code takes more time to give up on a
WM8805 or USB when there has been actual audio coming from that
interface. Only when the interface looses lock or becomes muted for
more than 2s does the code start looking at other inputs for valid
audio.

This mechanism also extends to the USB interface. With input_select ==
MOBO_SRC_UAC2 or MOBO_SRC_UAC1 the WM8805 is powered down and the USB
receiver owns the DAC's I2S interface. When the USB audio is muted or
halted, input_select is set to MOBO_SRC_NONE and the WM8805 is powered
up to start scanning its inputs. When it finds a valid one,
input_select is set to MOBO_SRC_SPDIF or MOBO_SRC_TOSLINK, and the I2S
interface is switched over to the WM8805. And when the WM8805 is muted
or looses link, the I2S is once again handed over to the USB system.
At first, with input_select == MOBO_SRC_NONE, this is done to mute the
DAC.

If this project is of interest to you, please let me know! I hope to
see you at either:
Audio Widget mailing list: audio-...@googlegroups.com
https://www.facebook.com/henryaudio
bo...@henryaudio.com