What is the maximum bits-per-symbol possible using FSK on telephone devices?
in [DSP]
|
From: Clay on 19 Dec 2009 16:35 On Dec 18, 11:20 pm, "steveu" <ste...(a)coppice.org> wrote: > >Clay wrote: > > >> On Dec 18, 12:06 pm, Vladimir Vassilevsky <nos...(a)nowhere.com> wrote: > > >>>Besides, there is no good way to achieve simulcast coverage with > 4-FSK. > > >> Basically data packets are built with a time to transmit > >> tag in them at the paging terminal and the packets are distributed to > >> the transmitters. The smart transmitter emits the scheduled paging > >> data at the right time. > > >Timing isn't a very big problem. The problem is interferrence between > >the transmitters. With 2-FSK, you can shift the carrier frequencies of > >transmitters by +/- 0.5 x baud rate. Then the mutual interference will > >be averaged out per duration of one bit, and everything works great. > >With 4-PSK, you can't do that. Which makes simulcast networking pretty > >much impossible. > > >> At one paging equipment manufacturer I worked on both the encoders and > >> then the protocol monitors and pager intercept devices. I used > >> Moto56309 DSPs for those projects. The company was already using that > >> processor and its predessor (56001 and 56002) in its product line, so > >> were very comfortable with manufacturing. I certainly enjoyed the > >> available horsepower and 24 bit depth! The specialized equipment was > >> not very price sensitive, so $30 or more for a DSP was not an issue. > > >I developed paging and PMR stuff. We had to do everything by i8051 and > >M68HC11. Analog circuitry and tons of assembly code. When ADSP-21xx > >became available, that was great relief. > > Its interesting that so many people here have worked on paging equipment, > when it was always such a very niche business, with very few DSP people > involved. > > Yours, another who developed paging systems, > Steve- Hide quoted text - > > - Show quoted text - Paging was the hot thing back during the 80s That's when I go into it and learned DSP. I think a few of us "old farts" go into it that way. And from the late 80s to the 90s, DSPs proved to be a very effective way to handle lots of stuff in a paging terminal. Clay
From: Steve Pope on 19 Dec 2009 17:53 Clay <clay(a)claysturner.com> wrote: >Paging was the hot thing back during the 80s That's when I go into it >and learned DSP. I think a few of us "old farts" go into it that way. Never worked on paging systems. I got into it from the digital audio angle, in the late 70's. Steve
From: Green Xenon on 13 Jan 2010 13:40 >Jerry Avins <jya(a)ieee.org> wrote: >(snip) > >> Thanks. > >> You wrote in another post that 1 symbol/sec satisfies you. Since >> sampling for a second theoretically yields 1 Hz resolution, that's a lot >> of frequencies in the telephone voice band. Using a spacing of 2 Hz to >> allow for less-than-ideal conditions, and taking the band to be 300 to >> 8,000 Hz, you can see that it is trivially simple to get 3,850 bits in a >> symbol provided you can synchronize properly. Now, 2^3850 is a big >> number, nearly 10^1159. What will you do with an alphabet of that many >> letters? > >That should be 4kHz. > >Also, the bits/symbol is log2(number of possible symbols), so log2(1850). > >Half the bandwidth might be enough to recover the clock. That is, >to know when the symbol transitions occur. There needs to be enough >(or sufficient) frequency changes such that the receiver can detect >where they occur, and not lose sync. > >A more efficient way is to use multiple BFSK carriers not overlapping >in frequency space. Then you do get a large number of bits. > >-- glen > According to http://www.motionnet.com/calculator/ log2(1850)=10.8533095554037 10.8533095554037 rounds off to 11. Does this mean the max amount of bits-per-symbol using FSK on a phone line is 11? What if I'm using a hypothetical FSK modem that does not require additional power supply other than the electricity provided by the phone line itself? Then what is the maximum bits-per-symbol possible?
From: Green Xenon on 13 Jan 2010 21:46 >Green Xenon wrote: >>> Jerry Avins <jya(a)ieee.org> wrote: >>> (snip) >>> >>>> Thanks. >>>> You wrote in another post that 1 symbol/sec satisfies you. Since >>>> sampling for a second theoretically yields 1 Hz resolution, that's a >> lot >>>> of frequencies in the telephone voice band. Using a spacing of 2 Hz to >>>> allow for less-than-ideal conditions, and taking the band to be 300 to >>>> 8,000 Hz, you can see that it is trivially simple to get 3,850 bits in >> a >>>> symbol provided you can synchronize properly. Now, 2^3850 is a big >>>> number, nearly 10^1159. What will you do with an alphabet of that many >>>> letters? >>> That should be 4kHz. >>> >>> Also, the bits/symbol is log2(number of possible symbols), so >> log2(1850). >>> Half the bandwidth might be enough to recover the clock. That is, >>> to know when the symbol transitions occur. There needs to be enough >>> (or sufficient) frequency changes such that the receiver can detect >>> where they occur, and not lose sync. >>> >>> A more efficient way is to use multiple BFSK carriers not overlapping >>> in frequency space. Then you do get a large number of bits. >>> >>> -- glen >>> >> >> According to http://www.motionnet.com/calculator/ >> log2(1850)=10.8533095554037 >> >> 10.8533095554037 rounds off to 11. > >You can't fit 11 gallons into a 10.8533095554037-gallon can. Round down. > >> Does this mean the max amount of bits-per-symbol using FSK on a phone line >> is 11? >> >> What if I'm using a hypothetical FSK modem that does not require >> additional power supply other than the electricity provided by the phone >> line itself? Then what is the maximum bits-per-symbol possible? > >Quote properly. You attribute some of Glenn's words to me. Glenn is >right about the bandwidth, but we make different assumptions about >counting bits. I assumed that every FFT bin could be counted as a bit, >independently of the others, analogously to a bundle of wires which can >be energized in any combination. If you redo my calculation with the >proper upper frequency and use half of that number to allow clock >extraction, you still get a lot of bits through. Just as there's no free >lunch, it's possible to do things in unnecessarily complicated ways >without losing capability. Ask Rube Goldberg. > >Jerry >-- >Engineering is the art of making what you want from things you can get. >����������������������������������������������������������������������� I posted the message using http://www.dsprelated.com/compdsppost.php?id=122814 replying to Glenn but the website instead quoted you. Sorry about that. Anyways, so 10-bits-per-symbol is the max in the scenario I describe? As asked before, what if the modems don't use any power supply other than the phone line electric current, what would be the max bits-per-symbol, then?
From: Green Xenon on 14 Jan 2010 15:04
>Green Xenon wrote: >>> Green Xenon wrote: >>>>> Jerry Avins <jya(a)ieee.org> wrote: >>>>> (snip) >>>>> >>>>>> Thanks. >>>>>> You wrote in another post that 1 symbol/sec satisfies you. Since >>>>>> sampling for a second theoretically yields 1 Hz resolution, that's a >>>> lot >>>>>> of frequencies in the telephone voice band. Using a spacing of 2 Hz >> to >>>>>> allow for less-than-ideal conditions, and taking the band to be 300 >> to >>>>>> 8,000 Hz, you can see that it is trivially simple to get 3,850 bits >> in >>>> a >>>>>> symbol provided you can synchronize properly. Now, 2^3850 is a big >>>>>> number, nearly 10^1159. What will you do with an alphabet of that >> many >>>>>> letters? >>>>> That should be 4kHz. >>>>> >>>>> Also, the bits/symbol is log2(number of possible symbols), so >>>> log2(1850). >>>>> Half the bandwidth might be enough to recover the clock. That is, >>>>> to know when the symbol transitions occur. There needs to be enough >>>>> (or sufficient) frequency changes such that the receiver can detect >>>>> where they occur, and not lose sync. >>>>> >>>>> A more efficient way is to use multiple BFSK carriers not overlapping >>>>> in frequency space. Then you do get a large number of bits. >>>>> >>>>> -- glen >>>>> >>>> According to http://www.motionnet.com/calculator/ >>>> log2(1850)=10.8533095554037 >>>> >>>> 10.8533095554037 rounds off to 11. >>> You can't fit 11 gallons into a 10.8533095554037-gallon can. Round down. >>> >>>> Does this mean the max amount of bits-per-symbol using FSK on a phone >> line >>>> is 11? >>>> >>>> What if I'm using a hypothetical FSK modem that does not require >>>> additional power supply other than the electricity provided by the >> phone >>>> line itself? Then what is the maximum bits-per-symbol possible? >>> Quote properly. You attribute some of Glenn's words to me. Glenn is >>> right about the bandwidth, but we make different assumptions about >>> counting bits. I assumed that every FFT bin could be counted as a bit, >>> independently of the others, analogously to a bundle of wires which can >>> be energized in any combination. If you redo my calculation with the >>> proper upper frequency and use half of that number to allow clock >>> extraction, you still get a lot of bits through. Just as there's no free >> >>> lunch, it's possible to do things in unnecessarily complicated ways >>> without losing capability. Ask Rube Goldberg. >>> >>> Jerry > >> I posted the message using >> http://www.dsprelated.com/compdsppost.php?id=122814 replying to Glenn but >> the website instead quoted you. Sorry about that. >> >> Anyways, so 10-bits-per-symbol is the max in the scenario I describe? As >> asked before, what if the modems don't use any power supply other than the >> phone line electric current, what would be the max bits-per-symbol, then? > >The source of the modem's power doesn't affect the theoretical bit rate. >The actual bit rate will depend on circuit design. The limit is set by >thermodynamics, but we're not close to being limited by that. > >The phone line is capable of 64 Kb/sec in theory and 56 Kb/sec in >practice. If you send a symbol per second, you can have 56 Kb/symbol. If >you send a symbol per hour at 64 Kb/sec, you get 3.36 megabytes/symbol. > >Jerry >-- >Engineering is the art of making what you want from things you can get. >����������������������������������������������������������������������� > Really? It's possible to convey 56,000-bits-per-symbol on a phone line using FSK [assuming a baud of 1-symbol-per-second]? Isn't their a limit to the amount of bits-per-symbol regardless of the amount of symbols per second? From what your saying, it should be theoretically-possible [using FSK] to send a Graham's-number-of-bits-per-symbol on a phone line if the symbol rate is low enough. Graham's number is one extremely large number! |