Using the Hammerfall HDSP on Linux

RME's Hammerfall HDSP sound card is a powerful, professional-level sound
card used in sound studios worldwide. It also works under Linux using
the ALSA Project drivers. Your average consumer sound card, for instance,
a Sound Blaster Live!, is designed around outputting sounds to a single
set of speakers, be it a stereo setup or surround sound. Sound
input, such as from a microphone, is an unconnected function done by the
rare program that needs it. The HDSP on the other hand is designed as a
sound I/O device. It has inputs and outputs, and you can route
sound between them arbitrarily. It's like the difference between a NIC and
a router.
Sure a NIC is useful, but if you want to run a network, or sound studio,
you need a full-fledged router.

So what exactly makes the HDSP special? First of all, it actually comes in
two pieces, a PCI or PCMCIA host card and an external sound I/O module. The
card and module are connected by a 10-foot Firewire cable, though
the protocol used isn't actually Firewire. Two types of
sound modules are available, the Multiface and the Digiface. The Multiface
offers eight channels of analog I/O and digital SPDIF, as well as ADAT
in and out for a total of 20 channels of 24bit/96kHz audio. The Digiface
offers three ADAT in and outs and SPDIF for a total of 26 channels. With
this separated design you can buy, for instance, a single Multiface and
share it between your desktop and a laptop. One example of where this is
useful is in live DJing. You don't have to buy two complete sound cards,
which saves you money, and you also can test software on your desktop knowing
that the environment is the same as it will be on your laptop.

This article focuses on using the Multiface module with the
PCI host card. If you have different hardware, most of this article
should still be applicable, and wherever possible, I've included information
on the differences.

Architecture
Figure 1 shows how sound is routed inside the HDSP. At the
top are the inputs, Analog, ADAT and so on, with the channel numbers below. At
the bottom are the outputs, again with channel numbers. In between is
the Matrix Mixer, which does the actual sound routing. In the Matrix
Mixer, any input channel can be routed to any number of output channels,
each having a selectable gain. All routing is performed internally
in HDSP with zero host CPU load. This gives you total flexibility in how
your sound is routed, for instance, allowing you even to
have an analog input go directly to a digital SPDIF output with no software
in between and no load on the host CPU.

Figure 1. Routing Sound within the Hammerfall HDSP

The Playback box at the top actually refers to applications
playing sound. Say you want to play a simple stereo MP3
with xmms. The first two playback channels would be used for left
and right. These channels are mapped to channel numbers 26 and 27 and
then are treated like any other input. What's interesting about this
approach is that playback channels have the same flexibility as any
other input channel. You can, for instance, send a playback channel
to multiple hardware outputs at the same time and even mix a playback
channel with a hardware input to produce an output. Unlike all
other input channels, the playback channels have a master gain control
for each channel.

Finally, due to various implementation details, the HDSP works a little
differently with sample rates above 48kHz. Because the ADAT has eight channels
with a sample rate less than or equal to 48kHz and only four channels for
a sample rate greater than 48kHz, the card uses two modes to distinguish
between them, SingleSpeed and DoubleSpeed mode. In SingleSpeed mode, on the
Multiface there are 18 channels, input and output. In DoubleSpeed mode
this is reduced to 14. This means that the channel numbers used by the
application change. In SingleSpeed mode, SPDIF is on channels 16 and 17,
and in DoubleSpeed this changes to 12 and 13. Unfortunately, ALSA has no way
of informing applications that the number of channels is being reduced, so
you must close all applications using the HDSP before switching between
the two modes. However, the matrix mixer channel numbers do
not change.

Hardware Setup
Download and install the latest stable release of the ALSA drivers from
www.alsa-project.org. You need to compile and
install the HDSP module and the hdsploader program in alsa-tools. It's a
good idea to make sure the snd-hdsp module is loaded first at boot time, because it needs to reserve a few chunks of contiguous memory for DMA
transfers. If it can't reserve this memory, the card won't work. Loading
the HDSP module first, however, makes it the default sound device
in ALSA. If you intend to use another card for general sound I/O, which is highly
recommended, use the index parameter when you load the module. For
instance,
my modules.conf has these two lines at the top to load my HDSP first
and then my Sound Blaster Live!:

snd-hdsp index=1
snd-emu10k1 index=0

All the command-line examples using amixer in this article assume you have
your HDSP at index 1. Once that is finished run the hdsploader
program to load the correct firmware into the Multiface or Digiface. It's
best to make sure this is done every time the drivers are loaded, so add
it to your bootup scripts. If you unplug your Multiface or Digiface, you
also need to reload the snd-hdsp module and run hdsploader again
when you plug it back in.

Finally, you should add a few lines to your .asoundrc file
in /etc/asound.conf or your home directory. The HDSP operates in
non-interleaved mode, which many applications can't use. You also
usually want to work only on a subset of the many input and output
channels available. With asoundrc you can attach labels to different sets
of channels. So instead of referring to channels 16 and 17, you could
refer to hdsp_spdif. An in-depth look at asoundrc configuration is beyond
the scope of this article, but the following should be a good start for
using the Multiface. If you have a Digiface the labels will be wrong;
hdsp_analog will correspond to your first ADAT connector, hdsp_adat to
your second and hdsp_spdif will be wrong. See the asoundrc documentation
for details on configuring asoundrc properly in this case.

pcm.hdsp {
        type hw
        card 1
}

ctl.hdsp {
        type hw
        card 1
}

pcm.hdsp_analog {
        type plug
        ttable.0.0 1
        ttable.1.1 1
        ttable.2.2 1
        ttable.3.3 1
        ttable.4.4 1
        ttable.5.5 1
        ttable.6.6 1
        ttable.7.7 1
        slave.pcm hdsp
}

pcm.hdsp_adat {
        type plug
        ttable.0.8  1
        ttable.1.9  1
        ttable.2.10 1
        ttable.3.11 1
        ttable.4.12 1
        ttable.5.13 1
        ttable.6.14 1
        ttable.7.15 1
        slave.pcm hdsp
}

pcm.hdsp_spdif {
    type plug
    ttable.0.16 1
    ttable.1.17 1
    slave.pcm hdsp
}

Testing the HDSP—Playing and Recording Sound
Now that the modules are loaded, and hdsploader has been run we can try
to play a sound using aplay. Like any other ALSA driver, the HDSP driver
mutes all channels at startup, so first you need to set the gain control,
or volume, properly for your playback channels. One way to do this is
to simply run alsamixer -c 1. When the bars are at 50% the channel is
output with approximately 0dB gain. A more precise way is using
amixer directly. The gain is as follows:

    0 = -inf dB
32768 =    0 dB
65535 =   +6 dB

A simple way to set all channels to 0dB gain is to use the following script:

#!/bin/bash
for i in $(seq 1 18);do
	amixer -c 1 cset name=Chn,index=$i 32768
done

Now that you've done that, hook up your speakers to the first two analog
outputs, and try the following:

aplay -D hdsp_analog some_stereo_file.wav

If all goes well you should see something like this:

Playing WAVE 'some_stereo_file.wav' : Signed 16 bit 
    Little Endian, Rate 48000 Hz, Stereo

By default, playback channels are connected to their respective hardware
outputs; if this isn't true in your setup, you won't hear anything. If this
is the case, read the following section on configuring the matrix mixer.

You also can change the -D option to send audio to different channels. For
instance, using hdsp_adat would send the audio to the first two adat
channels; hdsp_spdif would go to the spdif outputs. All of this is
controlled by asoundrc.

If playing a sound worked, try recording a sound with arecord. This
time, hook up a sound source to your first two analog channels, and run
the following:

arecord -D hdsp_analog -f dat recording.wav

Press Crtl-C to halt recording. If everything worked, recording.wav should
be a stereo file at 48kHz. In this case, the matrix mixer has no effect
on what's recorded; the first channel recorded always corresponds to
the first hardware input and so on.

Switching between SingleSpeed and DoubleSpeed Mode—Using
96kHz
As mentioned above, all software using the HDSP must be closed before
switching between SingleSpeed and DoubleSpeed mode. But normally, software
sets the sample rate. So what now? Well, there is a setting related
to the AutoSync functionality (more on that later) that can do the job,
Sample Clock Source. I won't explain exactly how it works just yet, but
basically, by setting the internal clock to different frequencies, the card
automatically switches modes. For instance, to switch to DoubleSpeed
mode run:

amixer -c 1 cset numid=11 6

To switch back to SingleSpeed mode run:

amixer -c 1 cset numid=11 3

In both cases, don't forget to make sure that no application currently is
using the sound card.

The Matrix Mixer
Sound routing is the most powerful feature the
HDSP has to offer.
Unfortunately, because it is not a feature found on many sound cards, the
interface used to control it is not an ALSA standard. As of
ALSA version 0.9.3, the matrix mixer interface is somewhat of a hack and
is subject to change. It's possible to write values to it but not to read
them back. In any case, the syntax used is source,destination,gain, where
gain is as explained above, and source and destination are as follows:

For the Multiface:

input: 0-7 (analog), 16-23 (adat), 
      24-25 (spdif), 26-43 (playback)
output: 0-7 (analog), 16-23 (adat) 
        24-25 (spdif), 26-27 (line out)

For the Digiface:

input_source : 0-25  (physical channels), 
               26-51 (playback)
output_source : 0-25 (physical channels), 
               26-27 (line out)

The routing for the Multiface is also shown in Figure 1. By
default, the playback channels are routed to the hardware output channels
in order. All other possible connections are muted (0 or -inf dB gain).

To change an entry, use the following command:

amixer -c 1 cset numid=5 src,dest,gain

For instance, you could direct the SPDIF input to analog out with the
following:

amixer -c 1 cset numid=5 24,0,32768
amixer -c 1 cset numid=5 25,1,32768

When you're finished, run the commands again, but with 0 as the gain instead
of 32768. Don't forget that you can have multiple channels going to a
single output or multiple outputs from a single input. Any combination
is possible—use your imagination.

Using SPDIF
SPDIF on the HDSP Multiface is pretty straightforward; however, because
ADAT and optical SPDIF share the same optical connectors you need to
tweak a few mixer controls. The Digiface uses a similar setup as well,
though not having access to one, I do not know the exact details.
To enable optical input for the SPDIF, change the IEC958 Input
Connector setting:

numid=6,iface=PCM,name='IEC958 Input Connector'
; type=ENUMERATED,access=rw---,values=1,items=3
; Item #0 'ADAT1'
; Item #1 'Coaxial'
; Item #2 'Internal'
: values=2

On later versions of the driver, setting the value to Internal seems
to autoselect between optical (ADAT1) and coaxial, on others it does
nothing. I don't know what the behavior will be with your driver, so you
should check yourself.

Optical output is similar. By default, SPDIF output goes only to the
coaxial connectors, but when the IEC958 Output (also on ADAT1, numid 7)
is enabled, output goes to the optical connector as well.

The Sample Clock and AutoSync
Like all digital audio devices, the HDSP needs a clock source to
function. This can be from either the internal clock generator or
an external digital source, such as the ADAT or SPDIF inputs or
the Word Clock. This is important when recording from external digital
sources; if the clock is not synchronized properly, dropouts and clicks can
result. When the clock is coming from an internal source, the HDSP is said
to be operating in Master mode, and if the clock is from an external source,
Slave mode. As shown in Figure 2, the various clock sources
go into the AutoSync system, which then sends a single clock signal to
the clock users.

Figure 2. The AutoSync system decides which clock signal to use.

AutoSync decides which clock signal
to use. This is controlled by the Sample Clock Source setting and the
Preferred Sync Reference setting. The former, Sample Clock Source, determines if the HDSP operates in
Master or Slave mode. Its possible settings are as follows:

numid=11,iface=PCM,name='Sample Clock Source'
; type=ENUMERATED,access=rw---,values=1,items=7
; Item #0 'AutoSync'
; Item #1 'Internal 32.0 kHz'
; Item #2 'Internal 44.1 kHz'
; Item #3 'Internal 48.0 kHz'
; Item #4 'Internal 64.0 kHz'
; Item #5 'Internal 88.2 kHz'
; Item #6 'Internal 96.0 kHz'
: values=0

If you specify AutoSync, 0, the HDSP operates in slave mode and follows
an external clock source. In this mode, playing a sound
works only if the sample rate matches the external sample rate; otherwise,
you receive an error message. Also, if there isn't a valid
external clock available, the driver will not let you switch to AutoSync
mode. Any other setting will set the Internal Clock Source sample rate to
the specified frequency. Once the card is not in AutoSync mode, playing
a sound at a different sample rate to the current one changes the
current sample rate.

The other setting, Preferred Sync Reference, is useful when you have
multiple possible clock sources connected but need to specify which one
to use as the clock source. Suppose you had a consumer
CD player connected with SPDIF and an eight-track tape recorder connected
with ADAT to your HDSP. Almost no consumer CD players can operate in
slave mode and receive a clock signal. So the CD player will have to
be the master device. However, AutoSync normally tries to sync to
the ADAT first. What now? Well, by setting Preferred Sync Reference to
IEC958 (SPDIF), AutoSync will synchronize the clock to the CD player. The
tape recorder then can be synchronized though other means. The possible
settings of Preferred Sync Reference on the Multiface are as follows:

numid=13,iface=HWDEP,name='Preferred Sync Reference'
; type=ENUMERATED,access=rw---,values=1,items=4
; Item #0 'Word'
; Item #1 'ADAT Sync'
; Item #2 'IEC958'
; Item #3 'ADAT1'
: values=2

ADAT Sync and ADAT1 are different things. ADAT1 refers to the
optical ADAT connection, and ADAT Sync refers to the D-sub 9-pin ADAT Sync
connector. There are also a bunch of read-only mixer items that show
various aspects of synchronization, type amixer -c 1
contents to see
them all.

The most common Linux sound complaint is, “I'm
getting clicks and dropouts in my sound!”
This could be caused by a number of factors, such as PCI bus bandwidth,
nonconforming PCI cards, too-small block sizes or latency. If
possible, install the low-latency kernel patches or the 2.6 kernel if
it's available by the time this article is published, and make sure
your audio program is set to use real-time scheduling. You also can try
increasing the block size your program uses. For hardware difficulties
various PCI cards, such as some 3Com network cards and RAID controllers,
have been known not to follow the PCI bus specification and use more
bandwidth then they should, starving the HDSP of bandwidth. You also might
need to change the PCI latency timer settings; for information on
this see www-106.ibm.com/developerworks/library/l-hw2.html. Some
motherboards also are known to have poorly performing PCI buses. RME's
Web site is quite informative when it comes to hardware incompatibility.
I recommend looking at the support section, both after and before
buying a Hammerfall.

Final Word
Hopefully, this article has helped you use your Hammerfall and has saved
you some time and trouble figuring out how it works. At this point,
you might be wondering “What now?” If so, try out some of the following
open-source software with your HDSP:

• Ardour (ardour.sf.net): a multichannel hard disk
  recorder (HDR) and digital audio workstation (DAW).
• Jack (jackit.sf.net): a low-latency audio server
  for connecting audio applications together.
• Ardour (ardour.sf.net): a multichannel hard disk
  recorder (HDR) and digital audio workstation (DAW).
• Pd (www-crca.ucsd.edu/~msp/software.html): a
  real-time graphical programming environment for audio and graphical
  processing.

Peter Todd has been using Linux since he was 14. He has a part-time
job as the head techie of a small Linux-based sound studio and is
greatly indebted to his boss for giving him the time and equipment to
write this article. When he's not working he attends Wexford Collegiate
where he currently is studying ceramics and graphics design.