Lo-Fi Phones and the Future 228
bossanovalithium writes "Back in 1936 — 74 years ago — boffins accepted that about 3.3Khz was the accepted frequency that telephone calls are going to run on and it's been like that, generally, ever since. Call quality is reasonable but leaves a lot to be desired. Think calls from Skype to Skype where quality is often crystal clear." It's crazy to me that (for people with decent mics at least) Ventrillo sounds better than corporate conference calls.
guess what (Score:4, Informative)
I live in 3rd world country and our major cellphone networks support hd-voice codecs.
503. (Score:2, Informative)
hmm, well it is running IIS...
Re:Pardon me, but... (Score:3, Informative)
Coral Cache .... (Score:5, Informative)
Just click here [nyud.net] and avoid the Slashdotting...
Bandwidth not Frequency (Score:5, Informative)
I can't even read the referenced article but I can tell you the phase ""Back in 1936 — 74 years ago — boffins accepted that about 3.3Khz was the accepted frequency that telephone calls are going to run on" is totally wrong.
What they meant to say was that the relevant bandwidth for understanding speech would be from 100Hz to 3.4kHz. Making the required bandwidth be 3.3Khz.
Ignore the slashvertisement for XConnect (Score:5, Informative)
the problem has been solved yet not been implemented widely. It's called ENUM and freely available and open. No need for proprietary XConnect stuff to implement this functionality, it's based off DNS and thus already has a widely available penetration. All people (and large corporations) need to do is actually use it.
Re:You think talking on an iPad is wrong (Score:1, Informative)
The iPad is very expensive as a phone, by any standard. And it's not even a video phone.
You don't need a $500 device to send and receive voice.
Right on the spot (Score:5, Informative)
I was pondering this exact stuff just today at work, since a phone call sounded kinda crappy, barely acceptable until I needed to involve 2 more people and put it on speakerphone, it became so bad we had to give up. I dropped the phone call, switched to skype, and damn what a big difference. The crappiness of POTS is ridiculous indeed, and although I see the need for compatibility, it can't die soon enough.
By the way, if you like Ventrilo, try Mumble [sourceforge.net], which, apart from being free and open source, which can't hurt according to the /. crowd, has really awesome sound quality, and you can setup your own private instance in minutes. Plus, for the MMO crowds, it has extremely low latency, awesome echo echo echo echo cancellation and built-in auto volume normalization (helpful when That Loud Guy Without Headphones keeps pressing his PTT and everyone's in pain)
It's a real problem. (Score:4, Informative)
While bandwidth is low, that's not the big problem. Quality is really hard to fix over networks with time jitter. Which is why VoIP and cell phone voice quality frequently suck. The best phone audio today is from an ISDN phone to an ISDN phone - end to end uncompressed full duplex digital with hard bit timing synchronization. (ISDN voice never caught on in the US, but it's widely used in some European countries.)
Wire-line telephony is 8 bits sampled at 8KHz, so the highest potential bandwidth is 4KHz. Compare CD audio, 16 bits sampled at 44.1 KHz per channel. Cell phones are worse; they're usually compressed down to 9600 baud or so. There are some high-end video conferencing systems with higher-bandwidth audio, but they're rare.
Re:Bandwidth not Frequency (Score:4, Informative)
Actually the bandwidth is 3.1kHz, running from 300Hz to 3.4kHz. This is the range of frequencies that conveys the most information relevant to intelligibility. Anything else makes it easier to recognise the speaker but doesn't make it easier to understand them and can make it harder to understand in noisy environments.
If you low-pass filter speech below 3.4kHz then mostly you only lose the high frequencies in sibilants, but you also filter out a lot of background noise. If you're really interested you could set up a media player to play recorded speech through a tunable bandpass filter and see what you can filter out before the speech becomes hard to understand. Once you've got a feel for how the filters affect the intelligibility, try playing it back in a noisy environment (or mix in a recording of the inside of a car or something).
The 300-3400Hz filter is pretty standard in communications, and it crops up in all sorts of places. I wrote a software-defined radio app that defaults to 300-3400 but is easily tunable up to 15kHz for either low or highpass (although if you highpass filter at 15kHz you won't hear much). Occasionally I'll use it to roll off above about 2.8kHz to remove high-pitched squeaky noises from adjacent transmitters.
Obligatory screenshot: http://www.gjcp.net/~gordonjcp/lysdr.jpg [gjcp.net]
You can see the yellow strip representing the passband of the filter. The (fairly weak) signal shown doesn't really have any strong components much above about 2kHz, and I could reduce the noise by sliding the leftmost (lowpass) filter in a bit. A quick explanation of what you're looking at - that's a spectrum plot of a chunk of the 7MHz amateur band, with lower frequencies on the left as indicated by the scale at the bottom. Since on 7MHz we use lower sideband (LSB) [wikipedia.org], the higher audio frequencies correspond to lower RF frequencies (further away from the red tuning cursor).
WB-AMR is comming to your mobile (Score:3, Informative)
Hi!
In a year or two, most GSM/W-CDMA networks will be upgraded to WB-AMR codecs.
Orange is already using it in Moldova and London, others are testing.
It is marketed as High Definition Voice.
WB-AMR uses 16 kHz sampling instead of classic 8 kHz . Together with better voice compression,
higher quality of voice is using same capacity (say, 12.2 kbit/sec) as we use today.
Of course, PCM is out.
Both sides of connection must support WB-AMR, and everything in between as well,
so for few years it might not be available across different networks.
If one terminal can not use it any more (maybe due to handover to GSM cell not supporting WB-AMR),
fallback to AMR/EFR is made on both sides, using 64k/56k PCM inbetween.
Technology is avaialble for quite same time, but terminal vendors are slowing it down.
Some 20% of all terminals have to support it, otherwise it makes no sense for operator
to buy all SW needed to implement it network wide.
Funny: good old GSM will soon get higher voice quality as ISDN.
73
Re:It's a real problem. (Score:2, Informative)
It's also widely used for telephone interviews on TV/radio, due to the improved clarity.
Re:Latency? (Score:5, Informative)
I've been working with VoIP in enterprise environments for a little over a decade. Latency is indeed a real issue and has to be considered, however it's not as restrictive as you might think. Generally speaking, if your ping is 150ms round trip you will not be able to distinguish a delay during an audio conversation, unless you're in the same room with them. Latency up to 300ms round trip is generally considered acceptable.
Cell phone conversations may or may not utilize VoIP during some legs of their calls. If they do, it's not between the phone and the tower unless you're using Skype or some other 3rd-party application on the phone. There is a distinction between encoding/decoding analog voice and how the digital signal is transmitted; you cannot consider cell phone calls to be synonymous with VoIP even though they do share some characteristics.
While cell phones do have highly variable horsepower in the CPU, the encoding/decoding is handled in purpose-built hardware chipsets, not on the CPU. It's unlikely that the type or brand of phone has any but a negligible difference in latency. Most people do not notice the latency in cell-to-cell conversations, so it may be that you're more sensitive to it for some reason.
Another factor is that some of the widely-deployed audio codecs used to compress voice were built and tuned for English speakers. Those speaking very dissimilar languages, such as Mandarin, may find that audio quality is poorer even on the same equipment.
Lastly, there are defined codec standards for wideband audio. Cisco has been including them on all their phones for several years; I assume other VoIP manufacturers have as well but do not have personal knowledge. I found that some customers did not like using them, as they are accustomed to hearing some level of white noise in the background and are prone to misinterpret a period of silence as call disconnection. If you've ever asked "are you still there?", the clarity of the call was greater than you expected or, possibly, wanted. Even with normal quality codecs we've had to inject comfort noise for years.
Little of the above applies to video. That's a whole different story.
And there's a good reason for that (Score:3, Informative)
Even in the dawn of telephony, frequency response was a significant issue. Besides the poor quality of transducers, the lines themselves weren't very good. Twisted pairs would have been nice, but early telephone wasn't twisted to improve common-mode rejection, it was twisted to keep the pairs together. Common residential service used something approaching zip cord from about 1960 on, maybe earlier. This isn't even twisted. You wonder why your DSL service is so crappy? I wonder how it even works at all. 10Base-T would barf on 30 feet of straight-line zip cord, and there is a good chance your house has 60-80 feet of it from the pole to the NT1. My first ISDN service at home was a fiasco, with load coils and conditioners being ripped out and new cable strung from the street to the complex demarc.
Frequency response is not the same thing as bandwidth (though they are directly related), but for telephone a 300-3300Hz response is intelligible and manageable. Doubling it to 6500Hz doesn't do a whole lot except consume bandwidth and marginally improve intelligibility. If you want fidelity, well, 12,500Hz is a good start. A loty of people never heard the flyback transformer on their old TVs vibrate, but I can hear them loud and clear. That's 15,750Hz.
And AM radio can sound very, very good. AM in America has a theoretical response of 16KHz, but currently is restricted in the U.S. to 10.2KHz (since 1989) to accomodate more stations and reduced interference from distant stations. The BBC at one time sent good audio, and a few shortwave stations did, and old AM radios had great speakers because they sent pretty good audio back then. Reducing response is also a way to extend range, along with compression, limiting, and a few other tricks that degrade ausio quality greatly. But AM is now the province of talk and news, so it doesn't seem to matter. FM, of course, also uses those tricks, and the result is nasty sound quality. To a generation broguth up on 128kbps MP3s, this is not a great loss. I code my music for my players at 320K or any of the lossless formats. 128k sizzle drive me crazy. And most FM music stations use MP3s anyways, they are largely programmed nationally and delivered over a satellite link. Tragedy.
To ask for improved sound quality in telephone is to ask for some compromises - fewer conversations over a given link, fewer conversations per cell tower, more Internet bandwidth. I'm pretty sure none of the incumbents will bother, as this ultimately results in increased direct costs, and probably zero increased revenue. Skype, etc., play with the codec and give apparently better results, the emphasis on 'apparently'. There are some clever audio tricks that will give a more pleasing experience with very little increase in bandwidth. Maybe Android can play with the audio, but I bet Apple could care less. The ILECS, bah!
So, the legacy of telephony is an old one, and has left us with something that works, but not as well as it could. Just a few more dollars, and you could have better!
Re:Latency? (Score:4, Informative)
Latency of the call is highly variable, and dependent on two factors:
1. How much latency is in the network?
2. How much latency is introduced by the VoIP conversion itself?
I joined Vonage about 5 years ago. On my first ISP, I got a little over 3/4 second of latency on a really good ISP connection. This was annoying, but not enough to really make me want to spend two and a half times as much for a landline with a non-portable number. Eventually, Vonage went through a stretch of upgrades to the firmware on my adapter and the latency dropped to about 1/4 second (all but unnoticeable). However, I traveled a lot a couple of years later and found that hotel connections tend to have a lot more latency, so I got a cheap prepaid cell for when I was on the road. I settled down to a local job again and had a lot of trouble with my new ISP for a while, resulting in poor call quality and very high latency, then we got that straightened out and I was back to 1/4-second delay, which was pretty much the rule until my company issued me a cell phone with unlimited minutes, so I ditched my Vonage line because I didn't use it. But friends who have joined since have reported very low delays, almost unnoticeable, as long as their connections were good.
So the technology has improved, but you are still dependent on someone who gives you the better tech, and on a good Internet connection between you and the adapter on the other end where the call is bridged back to a POTS network.
However, landlines have a few features that people have a hard time giving up. Whether you are willing to pay for them is a different matter.
1. No need to manage power to a device. If the wires are up, the connection is the telco's responsibility.
2. Real, honest 911 with pretty much 100% accurate location awareness. Your tax dollars at work (which are a generous chunk of the difference between telco and VoIP).
3. "Feedback loop" (you can hear yourself talk in your earpiece). This helps regulate your volume, which is why people tend to talk louder into cellphones (they don't get that feedback).
4. No-delay talking. When telcos use VoIP, they use really high-end gear and fast networks to support it.
5. True DTMF support. This has gotten a little better, but VoIP for the most part can't carry DTMF tones to sufficient clarity, so your local VoIP adapter has to recognize an attempt at one and generate a fresh tone that your analog gear can recognize. Conversations with people can occasionally be interrupted by a "BEEP" as your VoIP adapter misidentifies a sound in their voice or the background as a DTMF tone and faithfully reproduces it, and you may get occasional complaints of the same issue on the other side . If it fails to reproduce when needed and you run a menu system, your customers will really hate traversing your menus.
The net result of all of that is, well, you get what you pay for. Telcos are expensive, but you are pretty much guaranteed a good call every time. Most of the gear you probably own was built to analog specifications, and the telcos are good at maintaining that spec.
For most of us, cell or VoIP is sufficient. We're OK with slight delays, a less-than-perfect reproduction of our voices, the occasional errant DTMF tone, etc.
If you run a business and you strongly feel that clear telephony is a vital part of your business, then it's probably worth paying for in your case, or at least paying for a REALLY good Internet connection and high-end VoIP gear, not consumer-grade stuff. Though you could always run one VoIP line for a while and see how it works out (just use it for less critical calls to start with).
engineered (Score:2, Informative)
The bandwidth was not "accepted". It was set by the engineers that design the first analog telephone systems. It is a compromise between the need to have very small bandwidth per channel (so you can multiplex a lot of channels, and send them on the expensive long-distance cable) and the need to understand what the other person is saying and also, very important, to recognize who that person is (large bandwidth is better). They made some tests and this is how they found the sweet spot.
Re:guess what (Score:1, Informative)
Here in Scotland, our TV definition has gone plaid.
Plaid? Scotland? Shirley they're Welsh!
Re:Latency? (Score:5, Informative)
An MIT student did his thesis on Voice vs. Data lantency on cell phones, you might be interested in his methodology and results:
Quality of Service Analysis for Audio over Cellular Voice Networks [mit.edu]
it's bandwidth, not frequency. (Score:3, Informative)
I bet boffins did. But they were wrong. Serves them right for being boffins, I suppose.
It's bandwidth, not frequency. In the USA, POTS (Plain Old Telephone Service) lines are 3 KHz, specifically 400 Hz to 3.4 KHz. 400 Hz is the low frequency, and that is way above the lowest tones in most voices; while 3.4 KHz, the highest frequency passed, is way below the highest tones in most voices. But the reason for the choice was this range provides very good intelligibility -- that is, ease of understanding -- for almost all voices, and at the time, wider bandwidth meant more expensive components multiplied by a huge, and growing, phone system.
Basically, many nuances of speech were foregone as a matter of financial triage.
Not Boffins, Evolution! (Score:3, Informative)
Boffins did not decide this. It is a consequence of the evolution of human speech, both vocalization and hearing. The part of our speech that encodes content is the part that telephones have been engineered to convey. It's actually less than the 300-3400 Hz band, there's a mostly-useless band segment between the low frequencies and the higher ones that can be left out without much effect on speech recognition.
There have been a number of "hi-fi" schemes for telephony and bandwidth-limited radio. Some add bass, which is really cheap to do in terms of bandwidth because there is only a narrow band to be added and there's not much real information there at all so that you can compress the heck out of it and it still sounds like speech. AMBE+ does this on two-way radio, with a rather irritating synthetic bass. The other is to add more highs, and then you are going to mostly get more sibilants.
It is going to end up working better on the music-on-hold than it will on real voice.
Let's Make it Narrower! (Score:4, Informative)