Should you compress voice over your WLAN?

Voice aficionados from wireless vendors Symbol Technologies and SpectraLink agree that configuring wireless LANs for converged voice and data is a delicate balancing act.

The task requires adjusting packet lengths, possibly with voice compression, and jitter buffer sizes to maximize bandwidth efficiency. But these overhead-reducing steps must be weighed against the risk of losing large packets, which can degrade voice quality much more than small-packet loss.

Voice can have a significant impact on data throughput, particularly in 11M bit/sec 802.11b WLANs that become congested.

“If you are running five [toll-quality] phone calls, these pretty much saturate…an 802.11b access point,” says Rich Watson, director of telephony product marketing at Symbol.

Today, most WLAN systems run uncompressed, 64K bit/sec voice for best quality, using G.711 codecs for analog-to-digital conversion only.

“If I’m listening to music on hold, I want G.711 [encoding], so I can hear all the bass and treble sounds.  This involves bigger packets,” Watson says.

On the other hand, G.729, which compresses 64K bit/sec voice down to 8K bit/sec, is optimized for the human voice, with has a shorter range of frequencies. So G.729 can be used for voice conversations, while G.729-compressed music “will sound a little tinny,” Watson says. 

Ben Guderian, director of marketing at wireless phone handset maker SpectraLink, notes that most codecs sample and transmit packets every 20 milliseconds.

“You can send, instead, every 40 milliseconds. This cuts the number of packets you send in half and reduces overhead,” he says. Then again, he adds, if you lose that 40-msec chunk of communication, the negative impact on the conversation is greater.

Guderian notes that the codec you’ll need in your handset is up to the maker of your IP PBX. SpectraLink, which works closely with nearly all IP (and TDM) PBX makers to provide wireless phone extensions to their switches, supports “whatever codec the IP PBX maker specifies” in SpectraLink handsets.

Another consideration with voice compression is how much of the client device’s processing power it will consume. Symbol’s Watson estimates that compression can chew up 80%, making data applications very slow, if not offloaded onto a co-processor.

The Symbol PocketPC-based PDT 8146 handset, for example, comes “voice ready,” he says, with a special digital signal processor that offloads voice compression from the host processor.