Mixing and Surround Sound

MIXING & SURROUND SOUND

Introduction Digital Audio Mixing Dolby Pro-Logic Encoding Latency Versus Underflow Things I Learned

If you're like me and have a laptop and a Dolby Pro-Logic decoder for your television, you have probably tried hooking your computer up to the TV's audio system to see how all those games with the Dolby Pro-Logic logo at the start sound with three hundred watts of amplification and a room full of speakers.

So far I've been disappointed, the cut-scenes sound great, but as soon as the action starts I'm back to listening to the front speakers only. For my game I thought it would be really cool if I could create the effect of hearing someone shooting from behind, or when a ship finally explodes in a ball of flames the player could hear the sound of engines exploding from behind, followed the roar of the fireball rushing forward to surround the player in before the blackness of your final demise. A trip to the Dolby Labs World Wide Web site (www.dolby.com) introduced me to the theory behind Dolby Pro-Logic encoding and gave me the basis for adding support for it to my game.

Dolby Pro-Logic is a four channel system. Left, center, right and surround. The center channel is designed to anchor voices where the screen is. The left and right are designed for music and sound effects, and the surround channel is primarily for sound effects. Those four channels are encoded on the left and right stereo channels so that they can be decoded into four separate channels using some special circuitry inside a Dolby Pro-Logic amplifier.

The center channel in a Dolby Pro-Logic system is encoded by reducing the volume of the center channel by 3dB to maintain constant acoustic power across the front, and then putting that signal on both the left and right channels. To decode it, the amplifier takes the elements that are common to the left and right channels, filters them out, and feeds them through the center speaker. The neat thing about this design is that I don't need to do anything clever in my audio engine to get smooth panning from left to center to right and back since the Dolby Pro-Logic decoder takes care of this.

Encoding the surround channel is a little trickier, but now that I found the secret of Dolby's ingenious design, it was straightforward to implement. The formal Dolby specification says that the signal for the surround channel should be in the range of 100Hz to 7KHz and that the phase of the signal must be encoded as +90 degrees out of phase on the left channel and -90 degrees out of phase on the right, resulting in a 180 degree phase differential between the surround components of the left and right channels. The 90 degree phase shifts help mitigate cancellation effects, which we'll talk about in a moment.

Through experimentation I found that as long as the signal is 180 degrees out of phase between the left and right channels, it doesn't matter whether there is that +90 and -90 degree phase shift or not. Doing a +/- 90 degree phase shift in software would be tricky. However, doing a 180 degree phase shift is easily achieved by negating the sample before mixing it with one of the output channels. To encode a WAV file to play through the surround speakers, mix the sample normally with the left channel, then NEG the sample and mix it with the right.

Being a fan of big explosions, I wanted to be able to mix a track into more than one output channel so that an effect could fill the room from the side and surround speakers simultaneously. The first time I tried this, I discovered the joy of signal cancellation. When I mixed my audio for left, right, or surround individually it worked great, left and surround worked, but for right and surround, my effect came out the left speaker, the right and surround speakers stayed silent. About half a second of thought brought the realization that by adding the negated sample to the right channel, I had inadvertently cancelled out the original sample that I had added into the right channel.

I had to find a way to avoid this cancellation effect. Using a separate track for the surround channel was a ineffective since all tracks start at buffer boundaries so even a separate surround track would cause cancellation. With a little experimentation with my Dolby Pro-Logic decoder I discovered that if I mixed the surround signal as little as one sample later than the normal left and right signals (a 1/11000 second delay at 11KHz playback!), that it could pick out the surround signal.

Since most surround decoders generate delays of several milliseconds for the surround channel anyway, this delay is unnoticeable. My initial implementation took advantage of this, but I soon discovered that with only one or two sample difference, the output suffered from signal leakage between the left, right and surround channels. A minor architecture change and a complete rewrite of the assembler routines later, and I could mix the surround channel an arbitrary number of samples later than the left and right channels.

I settled on mixing the surround data 10 samples late although I found that 3 samples was adequate to obtain good separation between channels at 11KHz playback. Listing 5 shows two of the mixing functions in assembler along with their C declarations.

One thing I found playing with these routines was they don't always do what I expected. For instance, I expected the playback to come through the center and surround speakers when I played a loud effect on the surround channel, as well as on the left and right channels. In fact, it comes out the left, right and surround signals, while the center stays silent. This is great for big explosions since many surround amplifiers have the bulk of their power on the left and right channels. By lowering the volume of the effect on the surround channel, I caused the sound to converge from the left and right speakers to the center speaker.


[In the interest of conserving editorial space, code listings are available for download here.]