mb@sparrms.ists.ca (Mike Bell) (12/11/90)
>In article <136548@pyramid.pyramid.com> lstowell@pyrnova.pyramid.com (Lon Stowell) writes: >> In article <5435@navy19.UUCP> benyukhi@motcid.UUCP (Ed Benyukhis) writes: >> >57 Kbps on the voice grade line that is band limited to 3.4 Khz is contrary >> >to both Shannon and Nyquist rules. And so on to a discussion based on dubious assumptions... My book on information theory gives Shannon's result as: Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) Bandwidth for telephony ~= 3.4KHz Telebit, in their literature for the T2500 Modem, claim "up to" 18000bps in PEP mode without compression. To achieve this, the S/N power ratio would have to be >15.8dB. Anybody know what Signal/Noise ratios are likely to be found on normal phone lines? This would give an indication of the practical limit to modem speeds. NB. If the information source contains significant redundancy, (ie. is non-random) then far higher *apparent* rates are possible.
wolfgang@wsrcc.com (Wolfgang S. Rupprecht) (12/12/90)
mb@sparrms.ists.ca (Mike Bell) writes: >>In article <136548@pyramid.pyramid.com> lstowell@pyrnova.pyramid.com (Lon Stowell) writes: >>> In article <5435@navy19.UUCP> benyukhi@motcid.UUCP (Ed Benyukhis) writes: >>> >57 Kbps on the voice grade line that is band limited to 3.4 Khz is contrary >>> >to both Shannon and Nyquist rules. >And so on to a discussion based on dubious assumptions... >My book on information theory gives Shannon's result as: > Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) >Bandwidth for telephony ~= 3.4KHz >Telebit, in their literature for the T2500 Modem, claim "up to" 18000bps >in PEP mode without compression. To achieve this, the S/N power ratio would >have to be >15.8dB. Also, remember that the 3.4 kHz is the MINIMUM bandwidth the phone line will present to the modem. The sky is the limit. Don't folks regularly send 10 Mhz ethernet connections over (admittedly shorter sections of) these wires? I believe 64k baud ISDN links actually go from the CO to your jack over these same wires. The trick here is that it is fairly easy to get good freq responses out of short sections of twisted pair. Its just that long runs are killers. The specs are a worst case spec. In practice bandwidth is (hopefully) going to be much better. I wouldn't be surprised to hear that the 18k tb+ spec assumed that you connected the two modems with a 3ft section of phone patch cord. -wolfgang -- Wolfgang Rupprecht wolfgang@wsrcc.com (or) uunet!wsrcc!wolfgang Snail Mail Address: Box 6524, Alexandria, VA 22306-0524
witkowsk@optilink.UUCP (Dave Witkowski) (12/12/90)
From article <1990Dec10.162413.13959@sparrms.ists.ca>, by mb@sparrms.ists.ca (Mike Bell): > My book on information theory gives Shannon's result as: > > Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) > > Anybody know what Signal/Noise ratios are likely to be found on > normal phone lines? This would give an indication of the practical > limit to modem speeds. > Idle channel noise per BellCore TRY-0064 is: 20dBrnC for C-message weighting 39dBrn for 3-KHz flat response uucp: {pyramid, uunet} !optilink!witkowsk internet: optilink!witkowsk@uunet.uu.net "Shoot the lawyers! More skin on HBO!" - Leonard H. Putgrass
gsteckel@vergil.East.Sun.COM (Geoff Steckel - Sun BOS Hardware) (12/12/90)
In article <1990Dec11.185205.13752@wsrcc.com> wolfgang@wsrcc.com (Wolfgang S. Rupprecht) writes: > >Also, remember that the 3.4 kHz is the MINIMUM bandwidth the phone >line will present to the modem. The sky is the limit. Don't folks >regularly send 10 Mhz ethernet connections over (admittedly shorter >sections of) these wires? I believe 64k baud ISDN links actually go >from the CO to your jack over these same wires. The trick here is >that it is fairly easy to get good freq responses out of short >sections of twisted pair. Not quite true. I refer people to the experts on comp.dcom.telecom, but the 3.4 KHz is the limit for which the telco people minimize dispersion and maintain a reasonable frequency response through amplifiers, multiplexors, and all the other interesting equipment between your modem and the other one. With the same twisted pair, but different compensating networks, amplifiers, etc, at intermediate points in the circuit, you can get pretty much any response you want. The telco can also make your circuit pretty quiet. A twisted pair, per se, can be used into the megahertz if you know a LOT about its properties. ISDN works (at 176 KB/s) because the engineers worked very hard to overcome two problems: dispersion and attenuation. The subtler part is dispersion: differential propagation delay versus frequency. Messes the heck out of pulses, and really destroys complex modulation schemes. Within limits, however, the dispersion and frequency dependent attenuation are predictable for a copper loop to the Central Office, and this information is used to `pre-distort' the ISDN loop waveforms and to recover the data at the far end. T1 (1.544 MB/s) also uses twisted pair - with sophisticated phase, amplitude, and frequency compensation on the transmitted and received signal. Once at the central office, ISDN uses all-digital transmission to the terminating central office, which then uses the twisted pair loop to the user. These techniques of precompensation and recovery are used by some `analog' modems as well, but the possibility of FDM or TDM links (with attendant brick wall anti-aliasing filters) in the transmission path limits their usefulness. 14400 B/s with V32bis represents a triumph of engineering over ugly limitations. geoff steckel (gwes@wjh12.harvard.EDU) (...!husc6!wjh12!omnivore!gws) Disclaimer: I am not affiliated with Sun Microsystems, despite the From: line. This posting is entirely the author's responsibility.
floyd@ims.alaska.edu (Floyd Davidson) (12/12/90)
In article <3599@jaytee.East.Sun.COM> gsteckel@east.sun.com (Geoff Steckel - Sun BOS Hardware) writes: >In article <1990Dec11.185205.13752@wsrcc.com> wolfgang@wsrcc.com (Wolfgang S. Rupprecht) writes: >> >>Also, remember that the 3.4 kHz is the MINIMUM bandwidth the phone >>line will present to the modem. The sky is the limit. Don't folks >>regularly send 10 Mhz ethernet connections over (admittedly shorter >>sections of) these wires? I believe 64k baud ISDN links actually go >>from the CO to your jack over these same wires. The trick here is >>that it is fairly easy to get good freq responses out of short >>sections of twisted pair. > >Not quite true. I refer people to the experts on comp.dcom.telecom, >but the 3.4 KHz is the limit for which the telco people minimize dispersion >and maintain a reasonable frequency response through amplifiers, >multiplexors, and all the other interesting equipment between your modem and >the other one. With the same twisted pair, but different compensating >networks, amplifiers, etc, at intermediate points in the circuit, you >can get pretty much any response you want. The telco can also make your >circuit pretty quiet. > >A twisted pair, per se, can be used into the megahertz if you know a >LOT about its properties. ISDN works (at 176 KB/s) because the engineers >worked very hard to overcome two problems: dispersion and attenuation. >The subtler part is dispersion: differential propagation delay versus >frequency. Messes the heck out of pulses, and really destroys complex >modulation schemes. Within limits, however, the dispersion and frequency >dependent attenuation are predictable for a copper loop to the Central >Office, and this information is used to `pre-distort' the ISDN loop >waveforms and to recover the data at the far end. > >T1 (1.544 MB/s) also uses twisted pair - with sophisticated phase, >amplitude, and frequency compensation on the transmitted and received >signal. > >Once at the central office, ISDN uses all-digital transmission to the >terminating central office, which then uses the twisted pair loop to >the user. > >These techniques of precompensation and recovery are used by some `analog' >modems as well, but the possibility of FDM or TDM links (with attendant >brick wall anti-aliasing filters) in the transmission path limits their >usefulness. 14400 B/s with V32bis represents a triumph of engineering >over ugly limitations. What Geoff is saying is precisely correct, but I'd like to add a couple things that may make it easier to see what he means. T1 and ISDN are *digital*, just like your RS232 line. Your modem on the other hand encodes that digital signal into an analog transmission system. Two very different things and do not compare directly. When your 3.4 Khz voice line is digitalized and put on a T1 stream it takes up 56kbps. (It is done at a 64Kbps rate, but 8Kbps are used for other than voice data.) The absolute max bandwidth for an analog transmission over a telephone system is designed to be 4Khz, but in fact you can forget about using the upper end or the lower end of that spectrum. Most lines will have much less that 3.4Khz bandwidth. We only test the freq response from 400Hz to 2800 Hz! If you have ISDN or a T1 stream then each bit will be transmitted from one end of your connection to the other, using whatever bandwidth you are paying for. If you have an analog drop (a normal telephone voice line) you can put whatever you wish on it, but what comes out the other end is still going to be limited at best by a A-D conversion to 56Kbps. That would be assuming you are switching through a digital switch. On a local loop using an analog switch you might get better. But *any* type of carrier used to trunk your call to another switch will limit the bandwidth (4Khz for analog systems, 56Kbps for digital systems). There is an answer to "How do we get more bandwidth?" though. That is exactly what ISDN is designed/supposed/going to do. > geoff steckel (gwes@wjh12.harvard.EDU) > (...!husc6!wjh12!omnivore!gws) >Disclaimer: I am not affiliated with Sun Microsystems, despite the From: line. >This posting is entirely the author's responsibility. Despite my address, I have no connection with the Institute of Marine Science at the U. of Alaska Fairbanks. Despite my .signature, I do test toll trunks for a living. Floyd -- Floyd L. Davidson floyd@hayes.ims.alaska.edu Salcha, AK 99714 paycheck connection to Alascom, Inc. When I speak for them, one of us will be *out* of business in a hurry.
floyd@ims.alaska.edu (Floyd Davidson) (12/12/90)
In article <4986@optilink.UUCP> witkowsk@optilink.UUCP (Dave Witkowski) writes: >From article <1990Dec10.162413.13959@sparrms.ists.ca>, by mb@sparrms.ists.ca (Mike Bell): >> My book on information theory gives Shannon's result as: >> >> Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) >> >> Anybody know what Signal/Noise ratios are likely to be found on >> normal phone lines? This would give an indication of the practical >> limit to modem speeds. >> > >Idle channel noise per BellCore TRY-0064 is: > >20dBrnC for C-message weighting >39dBrn for 3-KHz flat response > C message weighting matches your ear. But 3Khz flat will include very low frequency power line noise that won't affect your modem any more than it does your ear. Neither is a very good measure for modems. Also note that Dbrn is *not* a signal to noise power ratio. It is a measure of noise power. The bigger the number the more noise. I don't have Bellcore TRY-0064, but 20 DbrnC0 is a very quiet channel. My guess is that figure applies only to the local loop. End-to-End limits depend on how long the circuit is or how many carrier sections it goes through. For example a circuit greater than 100 miles can have up to 33 DbrnC0 ICN (Idle Channel Noise). Also satellite circuits are allowed higher noise and may have to be measured using a 1004Khz tone at 13 Db below test tone level to hold the channel up (vox). The tone is notched out, but the depth of the notch determines the *best* reading possible. ICN is not a good measure of real life performance of a modem on a circuit. Impulse noise, phase shifts, and gain shifts are the limiting factors on dial up lines. Floyd -- Floyd L. Davidson floyd@hayes.ims.alaska.edu Salcha, AK 99714 paycheck connection to Alascom, Inc. When I speak for them, one of us will be *out* of business in a hurry.
mb@sparrms.ists.ca (Mike Bell) (12/12/90)
In <1990Dec11.185205.13752@wsrcc.com> wolfgang@wsrcc.com (Wolfgang S. Rupprecht) writes: >>And so on to a discussion based on dubious assumptions... >>My book on information theory gives Shannon's result as: >> Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) >>Bandwidth for telephony ~= 3.4KHz >>Telebit, in their literature for the T2500 Modem, claim "up to" 18000bps >>in PEP mode without compression. To achieve this, the S/N power ratio would >>have to be >15.8dB. >I wouldn't be surprised to hear that the 18k tb+ spec assumed that you >connected the two modems with a 3ft section of phone patch cord. Telebit Trailblazer Plus brochure states: "Dial-up asynchronous connections at speeds of up to 19,200 bps with data compression, 18,000 bps without data compression". (I think the 19,200 limit is due to the Trailblazer supporting a maximum DTE speed of 19,200).
mb@sparrms.ists.ca (Mike Bell) (12/12/90)
In <4986@optilink.UUCP> witkowsk@optilink.UUCP (Dave Witkowski) writes: >> My book on information theory gives Shannon's result as: >> >> Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) >> >> Anybody know what Signal/Noise ratios are likely to be found on >> normal phone lines? This would give an indication of the practical >> limit to modem speeds. >> >Idle channel noise per BellCore TRY-0064 is: >20dBrnC for C-message weighting >39dBrn for 3-KHz flat response For those without calculators to hand, these suggest: 20db with 3k4Hz bandwidth -> 22640 bps (not sure what C-message means!) 39db with 3khz bandwidth -> 38870 bps
koch@motcid.UUCP (Clifton Koch) (12/13/90)
From article <1990Dec11.185205.13752@wsrcc.com>, by wolfgang@wsrcc.com (Wolfgang S. Rupprecht): > mb@sparrms.ists.ca (Mike Bell) writes: > > Also, remember that the 3.4 kHz is the MINIMUM bandwidth the phone > line will present to the modem. The sky is the limit. Don't folks > regularly send 10 Mhz ethernet connections over (admittedly shorter > sections of) these wires? I believe 64k baud ISDN links actually go > from the CO to your jack over these same wires. The trick here is > that it is fairly easy to get good freq responses out of short > sections of twisted pair. Its just that long runs are killers. The > specs are a worst case spec. In practice bandwidth is (hopefully) > going to be much better. We're talking about modems which have to used normal voice telephone systems to transfer data. ISDN is not the same, though it does use the same twisted pair to transmit the data, it is done in an entirely different manner. Ethernet is an entirely different matter, and though it can still use twisted pair, it is subject to a lot more constraints. The bandwidth is *not* limited to 3.4KHz by the wires. The bandwidth is limited due to the way the audio is trasmitted at the central office (phone company). The audio signal is sent by 12 bit samples compressed to 8 bits (by either Mu-law or A-law compansion, depending on the country) sent at 8,000 samples per second. Each audio channel is then multiplexed into a serial stream of 24 or 30 channels (again depending on the country). Nyquists theorm limits the bandwidth to 4Khz, though in practice it is limited to about 3.4KHz to avoid aliasing. -- ----------------------------------------------------------------------------- ... [uunet | mcdchg | gatech]!motcid!koch
floyd@ims.alaska.edu (Floyd Davidson) (12/13/90)
In article <1990Dec12.154139.20496@sparrms.ists.ca> mb@sparrms.ists.ca (Mike Bell) writes: >In <4986@optilink.UUCP> witkowsk@optilink.UUCP (Dave Witkowski) writes: > >>> My book on information theory gives Shannon's result as: >>> >>> Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) >>> >>> Anybody know what Signal/Noise ratios are likely to be found on >>> normal phone lines? This would give an indication of the practical >>> limit to modem speeds. >>> > >>Idle channel noise per BellCore TRY-0064 is: > >>20dBrnC for C-message weighting >>39dBrn for 3-KHz flat response > >For those without calculators to hand, these suggest: > 20db with 3k4Hz bandwidth -> 22640 bps (not sure what C-message means!) > 39db with 3khz bandwidth -> 38870 bps Scratch that, totally. Dbrn measures noise power, not signal to noise ratio. For example 20 DbrnC is less noise than 39 Dbrn. C message weighting is a filter that has approximately the same roll off as your hearing. Hence a 120 Hz power line harmonic can be quite strong and not read very high with C message weighting or bother you very much on a voice call. If anyone really cares (doubtful) send me email and I'll look it up and mail you some exact definitions. Floyd -- Floyd L. Davidson floyd@hayes.ims.alaska.edu Salcha, AK 99714 paycheck connection to Alascom, Inc. When I speak for them, one of us will be *out* of business in a hurry.
floyd@ims.alaska.edu (Floyd Davidson) (12/13/90)
In article <1990Dec10.162413.13959@sparrms.ists.ca> mb@sparrms.ists.ca (Mike Bell) writes: >>In article <136548@pyramid.pyramid.com> lstowell@pyrnova.pyramid.com (Lon Stowell) writes: >>> In article <5435@navy19.UUCP> benyukhi@motcid.UUCP (Ed Benyukhis) writes: >>> >57 Kbps on the voice grade line that is band limited to 3.4 Khz is contrary >>> >to both Shannon and Nyquist rules. > >And so on to a discussion based on dubious assumptions... > >My book on information theory gives Shannon's result as: > > Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) > >Bandwidth for telephony ~= 3.4KHz > >Telebit, in their literature for the T2500 Modem, claim "up to" 18000bps >in PEP mode without compression. To achieve this, the S/N power ratio would >have to be >15.8dB. > >Anybody know what Signal/Noise ratios are likely to be found on >normal phone lines? This would give an indication of the practical >limit to modem speeds. > >NB. If the information source contains significant redundancy, (ie. is >non-random) then far higher *apparent* rates are possible. I did a little research and looked it up. The old Ma Bell thought a reasonable S/N power ratio was 24 dB. However ( ain't there always a catch?), that is a static measurement. The numbers for impulse noise are 1 per minute over a 15 minute period with a threshold of 6 dB below data level. So calculate the "Channel Capacity" for a 24 dB S/N, and then figure on one hit per minute. (Numbers are from "Notes on the Network", ATT, 1980. They may have changed slightly since 1980, but I doubt it enough that I didn't bother comparing with the 1986 revision.) Also in another article 20 dBrnC was mentioned. That is maximum idle channel noise on a subscriber loop, as measured at the customer end. It does not include any connection past the line switcher. It is also the quietest portion of any connection you will ever dial up! Floyd -- Floyd L. Davidson floyd@hayes.ims.alaska.edu Salcha, AK 99714 paycheck connection to Alascom, Inc. When I speak for them, one of us will be *out* of business in a hurry.
mb@sparrms.ists.ca (Mike Bell) (12/13/90)
[This thread is getting complicated and it is no longer clear who is quoting what about whom... I find myself getting quoted as making the replies to my own posting... We hope we're not schizophrenic, but...] To summarise: Shannon gives: (note 0 noise = infinite capacity for >0 Bandwidth) ++++ Channel Capacity = Bandwidth * log2( 1 + SignalPower/NoisePower ) on telephone links involving digital transmission, we know that the maximum channel capacity is <=56Kbps (since this is the information flow rate of the digitized signal). On an analog link, a realistic Signal/Noise ratio (and thus the capacity limit imposed by the analog part of the link) is still open to bidders, since the dBrn figures turned out to be noise power figures, not S/N ratios. Any offers here? Nyquist still seems to be misunderstood/misquoted: my understanding is that the theorem states that: a signal whose highest frequency is F can be reconstructed non-ambiguously from samples taken at a sampling rate of 2F. (The samples, of course, will each require an infinite number of bits (I) to represent them exactly - giving rise to an information flow rate of 2FI. :-) [Please correct me: should that be a bandwidth limited signal?] Hopefully (once the analog S/N figures are known), we can lose all the discussion which says that "you can't do that because Nyquist and Shannon and ... say you can't", when that isn't what was said they said. [If there is a comp.dcom.telecom.standard.answers, a summary of all this should be in it...] -- Mike -- <mb@sparrms.ists.ca>
dricejb@drilex.UUCP (Craig Jackson drilex1) (12/14/90)
In article <1990Dec11.185205.13752@wsrcc.com> wolfgang@wsrcc.com (Wolfgang S. Rupprecht) writes: >Also, remember that the 3.4 kHz is the MINIMUM bandwidth the phone >line will present to the modem. The sky is the limit. Don't folks >regularly send 10 Mhz ethernet connections over (admittedly shorter >sections of) these wires? I believe 64k baud ISDN links actually go >from the CO to your jack over these same wires. The trick here is >that it is fairly easy to get good freq responses out of short >sections of twisted pair. Its just that long runs are killers. The >specs are a worst case spec. In practice bandwidth is (hopefully) >going to be much better. A dry local loop is indeed capable of much higher bandwidth--suitable for ISDN, etc. However, the nature of twisted pair is that it is capacitive, which tends to attenuate the highs. Long local loops (over 1 km or more) have inductance added, in the form of 'loading coils'. These basically kill ISDN, etc for that loop. Also, nearly all connections which are not completed within a single exchange go through some form of frequency-division multiplexed link (known as 'carrier' of various sorts). These do indeed go through filters. >I wouldn't be surprised to hear that the 18k tb+ spec assumed that you >connected the two modems with a 3ft section of phone patch cord. Naw.... if that were true, they'd have claimed 10Mbps... -- Craig Jackson dricejb@drilex.dri.mgh.com {bbn,axiom,redsox,atexnet,ka3ovk}!drilex!{dricej,dricejb}