>[ Forwarded message ] >Date: Thu, 16 Jan 1997 03:31:47 -0600 >From: George Wenzel >Subject: Re: X2 >To: Multiple recipients of list IAP > >Wanna know if 56k modems will work for you? Here is an emgineering >perspective on maximum line speed. > >These days with modern, high-quality, digital equipment, the common >perception is that the limiting factor is the copper wire at the end of >the circuit. While this is true if you are talking about a voice signal, >but not so true with modern modems. > >The primary limiting factor in an otherwise high-quality circuit is >the noise injected into the circuit at every A to D conversion. > >Modern modems use subtle signal changes to indicate the point on a two >dimensional grid that that data bit represents. It typically means several >actual bits of data. Slight distortion in amplitude or phase shift the signal >into a different grid (if the distortion is strong enough) causing the >data to be read in error. As modem speeds go up, sensitivity to these >slight distortions also goes up. > >The digital phone data is called PCM. Pulse Coded Modulation. > >At every point of PCM conversion, there is a type of noise called PCM >quantization noise. In effect, what PCM Q noise is is the rounding errors >that happen when a point on a signal is in between two of the 256 possible >digital values. There is some slight rounding on nearly every point encoded. > >Usually the impact of this rounding is only very slight, nearly impossible >to hear. However there are often two to six PCM conversions in a modern >phone call. The Q noise is cumulative, and at some threshold, it causes >errors with modern modems. > >Since Q noise is due to rounding of values while encoding a signal, the >effect of multiple layers of PCM is very predictable. Since the noise >is just a math function, it is just an issue of math as to how much >noise is generated on a given signal. It is important to understand >this, it is not a quality issue. No amount of money spent on quality >is going to change the essence of what causes the noise. > >A graphic example, I have $1.07 in my pocket. I am told to round that >amount to the nearest dollar. Whether I do it in my head, on a $3 >throw-away calculator, or on a $7,000,000 super computer, the answer >is always $1. The quantity of rounding "noise" is always $.07. > >While it is true that some PCM codecs are slightly more accurate than >others, it is the rounding errors that generate 95% of the noise. > >Ok, so what does this all mean to me? It means the number of PCM >conversions between you and the remote site is the hard limiting >factor on how fast your new modem will connect at and still have a >error-free connection. > >For a 14.4 connection, 6 PCM conversions is bearable by most modems. >For a 28k connection, 3 PCM conversions is bearable by most modems. >For a 33k connection, 2 PCM conversions is bearable by most modems. >For a 56k connection, 1 PCM conversion is bearable by most modems. > > >So, what causes different numbers of PCM conversions between me and >various parts of my local city? Well, if you use a digital switch >like most of us do, then it takes two PCM conversions to get from one >side to the other. If you phone company uses fiber to run to your >neighborhood, then depending on the equipment, they may be adding >two more PCM conversions. If the other end has fiber, then two more there. >Then there are those old analog switches that are still installed in >some places... depending on how they are patched into the digital network, >there are often additional PCM conversions there. > >Have you heard of a "slick" (actually it is spelled SLC), well, if you >have an SLC, then there are three PCM's between you and the switch. >If you live in a modern city with new fiber being run all over the place, >then it is quite possible that the other end is an SLC too. > >So why on earth do they put in so many PCM conversions if they are so bad? >The answer is that are not bad for voice calls, AND using analog >to run from the switch to the fiber means the 30 year veteran phone tech >can still clip his circa 1969 test set to one of the lines, and troubleshoot >it the same way he was taught back in 1967. Retraining is mostly eliminated. >Of course no one mentions that SLC equipment costs 4 times the price you can >get digital channel banks for. Channel banks fed by T1's have only 1 PCM >conversion between you and the switch. Why isn't it done this way? >I can give you a reason for each day of the week. None of the reasons >hold water from an engineering point of view. But if PUC regulation >limits rate increases to that needed to build the service, then spending >more on hardware building the service means the PUC will be more likely to >let you pass tarrifs for higher amounts.... you need only show that you are >spending the money. > >Another excuse for using SLC equipment is that they still have an old analog >switch in use, so the SLC is the cheapest manner with which to connect an >analog switch to a fiber trunk..... but is it really? Actually no. >For the price of a half dozen SLC installations, you can buy a new digital >switch with 10 to 100 times the capacity of the analog switch.... and never >have to buy expensive SLC equipment. > >Oh, and if they tell you that they have a new card for the SLC that doesn't >have as much PCM noise, do not believe them! Remember, a cray rounds numbers >to the same value that a cheap calculator can... how can a different card >change the fact that PCM is being used? The fact is I have seen these cards, >and they made no difference. The better cards have a higher-quality >analog section, sutable for handling somewhat longer copper wire runs, >but there is no effect on the PCM math. None, zip, zilch, nada. > >I could go on.... but hopefully you get the picture, and perhaps even learned >something. > > >My first engineering job was prototyping a fax modem. The next was to >prototype a T1 channel bank, with extended loop range on the T1 side. >The next was a Macintosh clone project, and then a multimedia >telephony appliance with every feature you can think of. Then I >connected my BBS to the Internet, and haven't had time to get a real job >ever since. > >George >-- >George Wenzel | There are no real limits... >Real/Time Communications | You are only prisoner of your own fear, >george@realtime.net | so fear nothing and you will see no limits. >--------------------------------------------------------------------------