Mobile VoIP in the context of mobile data capacity

When VoIP traffic began to grow in the late-90s, there was some worry that voice traffic would overwhelm the Internet.  Indeed at that time, voice traffic far exceeded total Internet capacity.  But the growth of Internet capacity has far outstripped the growth of voice minutes.  Today a substantial portion of long distance and international voice traffic goes via VoIP and increasingly, local traffic is moving to VoIP, yet VoIP represents only 1% of Internet traffic.  Voice traffic grew relatively slowly.  Total Internet capacity has expanded at ~100% per year during the 1990s and at greater than 50% per year since 2001..

So far, 3G mobile VoIP traffic is negligible and most mobile data traffic is confined to walled gardens.  But that's changing as mobile data capacity expands.  Today, laptops with USB 3G modems are driving 3G data revenues as people seek Internet access for their PCs.  Despite whatever Ts & Cs the operator may impose, what people want is open Internet access, i.e. a mobile dumb pipe.  And given the relatively vibrant competition in many mobile markets, increasing mobile capacity will lead to better and better "dumb pipe" offers.

The issues for successful 3G mobile VoIP are:

• capacity - if you have excess capacity, you don't need QoS.
• latency - round trip latencies must be below 250 ms to avoid a "push-to-talk" feeling.
• battery life - mobile VoIP may require some evolution to achieve the power efficiencies of the custom engineered protocols used for traditional mobile voice.
• devices - mobile VoIP will start on PCs, but must migrate to handsets to achieve volume.

Luckily battery life issues have been explored as part of implementing POC (push-to-talk over cellular) and should be well understood at this point.

The advent of HSUPA and EVDO Rev A radio technology brings latencies into the acceptable range.  Deployment of these technologies has begun.

That leaves capacity.  While there will be congestion during the "busy hour," daily traffic is cyclic.  During most of the day most cell sites have extra capacity.  And a 3G HSUPA modem handles 3-7 Mbps down and 1-2 Mbps up.  That's more than adequate headroom for VoIP.  In other words capacity is becoming available for most hours of the day.

Fixed VoIP took a decade to become pervasive and mobile VoIP will likely take as long, but the next 24 months should see early adoption flourish.

Comments

[Nik]

This part is interesting (well it's all interesting, but this part bugs me):

latency - round trip latencies must be below 250 ms to
avoid a "push-to-talk" feeling

and:

HSUPA and EVDO Rev A radio technology brings latencies
into the acceptable range

RTT on a transoceanic internet link is greater than 200 ms, maybe greater than 300 ms. (It can be even more, but that's when things aren't going well.) How much do low latency wireless connections add? How does the latency budget break down over the different legs of a Voip-over-mobile-data call?

I used VOIP over a Sprint high speed wireless card (err... I think this means EVDO Rev A) to call from my laptop in one Boston suburb to my house in another Boston suburb. My house has VoIP from Comcast. It worked with "demo quality".

I tried to call from the same laptop to a Verizon cellphone also in the boston suburbs - result was about the same, maybe a little worse.

Of course this is not data, it's just an anecdote. But my tentative conclusion is that voice over mobile data is one or two infrastructure turns away.

[brough]

Nik, Yes there could be issues with multi-leg calls! Especially if you are connecting wireless to VoIP with transcoding in between, as each stage then introduces measurable latency.

One question - is your Sprint high speed wireless card EVDO Rev 0 or EVDO Rev A. If it's Rev 0, then it doesn't matter what Sprint's latest tower provides, you will only get Rev 0 results. Rev 0 claimed 2.4mbps down and 144kbps up as theoretical max speeds. The actual speeds were more like 600k/60k in peak hours and 800k/90k in off-peak times with perhaps 340ms latency. The theoretical speeds for EVDO Rev A are 3.1mbps / 1.8mbps with less than 60ms latency. I've heard actual speeds on the order of 2 Mbps down and 400-500 kbps up. I haven't got anyone's field measurements of actual rev A latencies...

And remember, I only claimed we'd see early adopters in the next 24 months. :-)
I do think it will take a few more revs before this stuff crosses over to the main stream.

[Nik]

Thanks for the info. Ya know, I did actually miss the last sentence where you said 24 months - my brain was already huffing and puffing about latencies :^)

I did my Sprint experiment in early 2007 so I guess it was Rev 0. I don't have the card anymore, so I can't check.