MENG on MODEMS

The following information is more or less all my knowledge of modems collected since I bought my first modem in July 1994, it was a US Robotics 14.4Kbps Internal Sportster and has since departed the land of the living, thanks to lightning.
Disclaimer: I am not an expert in modems, so this information may be incorrect or out of date. Do whatever you wish with this information, but don't complain if you make an incorrect decision on the basis of whatever I write here.
If you wish to include this information with anything, all I ask is, tell em where you got it from :)

WHAT KINDS OF MODEM ARE THERE

There are many types of modems in use widely, thereV.32, V.32bis, V.34, V.34bis modems being used widely. More commonly known as 9600, 14400, 28800 and 33600 bps modems respectively. Recent additions include two flavours of 56Kbps modems, K56 Flex and X2. The standards have not been ratified and are not likely to be ratified until 1998/99. As such both are proprietary and non-standard. Rumour has it that the final standard will be a combination of the features of both K56flex and X2.

TELEPHONE LINE BANDWIDTH AND MODEM PERFORMANCE

The human voice which generally has a bandwidth range of 300Hz to 4000Hz. Telephone lines were, in general, designed for voice calls and therefore have bandwidths in this range. How much bandwidth a particular line has depends on:

DISTANCE

Distance from your phone to the telephone exchange (or switch if you are in the US). Basically the further the distance, the weaker the useful signal, signal strength is lost due to circuit reactance. Reactance is the combination of resistance, capacitance and inductance (see below for a basic description of these) of the wire, connectors and whatever equipment or device the signal passes through. Transmit a signal far enough and the signal becomes so weak that it becomes indistinguishable from the background noise !

An analogy is this: Imagine you are in a restaurant, the murmur of other people talking in the background represents the background noise. If you are close to your dining companions, you can hear them over the background noise. As you move further away, there will come a point where you cannot distinguish what they are saying from the background noise, unless you raise your voices. Not many modems are capable of doing the equivalent of raising voices and these are for leased circuits (permanently allocated circuit). Signal loss of any circuit is also dependent on the frequency (the pitch of a tone) of the signal. The reactance of the circuit determines the upper and lower frequency limits a wire will transmit. In the case of a voice telephone line lower limit can be 300-500 Hz and the upper limit 3000-4000 Hz.

Resistance - think of it like swimming up river against a current. This is an easy one to explain !
Capacitance - properly called Stray Capacitance. High frequency signals can travel across insulation of a wire to a parallel wire. Depends on the the frequency of the signal, length of the parallel wires, their physical proximity and the thickness of the insulation. Capacitance causes high frequency signal loss in this way.
Inductance - Electric currents create magnetic fields around a wire. The opposite is also true, magnetic fields generate electric currents in a wire. Under certain conditions, such as in a coil of wire, the magnetic field generated by a current, in turn generates a current opposing the original current. This degrades the signal strength of the original signal. This is also frequency dependent and the higher the freguency the higher the loss.

Note that both Stray Capacitance AND Inductance only affect high frequency signals and these two effects are the primary reason why circuits have an upper frequency limit. Only resistance affects low frequency signals and unless designed specifically to reduce low frequency signal strength circuits have no low frequency limit apart from that determined by the noise level.

NOISE

All electrical and electronic circuits have noise. Noise is random and is caused by random electron movements due to thermal excitation. All circuits have a baseline noise level (if the noise level is high, it is audible as a hiss) and is uniform throughout the length of the wire. Additional devices and equipment will add or subtract from the overall noise level. A high noise level means that weak signals cannot be distinguished from the noise. For example, in very simplistic terms, if a signal at 3500 Hz cannot be distinguished from noise but at 3400 Hz the signal is clearly distinguished, the upper limit of the bandwidth is about 3400 Hz. (Its actually more complex than this but this is as detailed as I want to go with this). The quality of a circuit is measured using the Signal to Noise (SN) ratio, that is the ratio of the signal strength compared to the noise level. A high signal to noise ratio means a good circuit. As explained above, as the circuit distance increases, signal strength drops. As noise levels are more or less constant (just like the restaurant background noise !) throughout the circuit, the SN Ratio drops.

EQUIPMENT

Modems may have features which can increase the bandwidth of a particular link. The effectiveness and availability of such features determines the actual bandwidth of a connection. Generally, the more expensive the modem, the more features it will have and the more usable bandwidth is available.

Some Phone companies use a devices called Digital Pair Gain Equipment. This device allows two or more voice channels to share one physical telephone wire pair. Typically DPGE are used when there are multiple phones needed in a single customer's premises. This is bad news to modem users if the DPG device splits the channels permanently. These are the cheap and therefore more common ones. The more expensive (and less common) ones allocate the channel dynamically. Why is DPGE bad for modems ? Two reasons, even with "good" DPGE any additional devices in the line will ADD NOISE AND DISTORTION. The bad DPGE HALVE your line bandwidth ! For example, lets say you had enough bandwidth to connect at 28.8Kbps, now you can only connect at 14.4Kbps or worse due to the increased noise floor and distortion introduced by the damned thing. Cheaper modems may not be able to handle it and may refuse to connect altogether ! The solution ? Just say NO to DPGE ! How do you know you are stuck with one ? The only sure way to tell is that suddenly your modem which used to give you 28.8K is giving you a lot less.

HOW MUCH BANDWIDTH DOES A MODEM NEED

V.34 Modems:
To
Connect 
at        Required Bandwidth
28800     3200 Hz    from 320Hz to 3520Hz
26400     3000 Hz    from 375Hz to 3375Hz
24000     2800 Hz    from 467Hz to 3262Hz
This means if the bandwidth of your telephone line drops a mere 400Hz from 3200 to 2800 Hz, your connect speed drops to 24000 bps.

I don't have the figures for V.34bis and 56K modems except that they need even more bandwidth.
All you physicists out there will be wondering, how does one send data at 28800 bps when the bandwidth is only 3000 odd Hz ? The quick answer is that the modem transmits waveforms (called Symbols) which represent more than 1 bit of data ... if I remember right V.34 transmits 8.4 bits per symbol and V.34bis 9.8 bits per symbol.
The modem starts off by testing the phone line (the funny noises you hear when your first connect). The modem tests the line for bandwidth, noise, distortion etc etc and decides on the best carrier frequency to use to minimise distortion and maximise signal to noise ratio. The modem also talks to the modem on the other end firstly to find a common protocol (modem language) both understand, for example, a V.32 modem and a V.34bis modem will talk V.32 to each other. They then negotiate error correction and compression protocols.

MODEMS : THE GOOD, THE BAD AND THE UGLY

Not all V.34 modems are the same. Just think about it, do you think a V.34 modem costing US$100 will perform as well as a V.34 modem costing US$500 ? Surely not ! Why ? You may ask, they are BOTH V.34 compliant ! Well, the V.34 standard did not make ALL its recommendations mandatory. So Mr Cheap Modem Manufacturer leaves out as many features as he can just to meet the MINIMUM V.34 compliance. That way he minimises production costs and can sell you the modem really really cheap. Mr Good Modem Manufacturer implements as many as he can, that means the modem costs more to manufacture to they don't come cheap. Cheap modems leave out features that can maximise bandwidth for any given line ... features which reduce the impact of line noise and distortion for instance. What does that mean ? That means a cheap modem may manage to squeeze only 2800 Hz bandwidth out of a phone line and so be limited to 24000 bps whereas a good modem will the full V.34 features may manage 3200 Hz (a mere 400 Hz more) and get 28800. Some of the very worst modems I have heard about are those internal modems which come with so-called, Internet Ready, PC's. Manufacturing costs for these disgusting things are probably in the US$20 range. If you have one of these, do yourself and your phone bill a favour and replace it ASAP. My personal favourite is the US Robotics Courier V.Everything (flashed with the X2 code). I have an external version for my desktop at home and a PCMCIA version for my laptop. There is a catch, the US$400 odd street price, but believe me, it will pay for itself within a year by saving you time and phone bills, not to mention the endless frustration of dropped connections.
SOME V.34 modems (like the Courier) can transmit and receive at asymmetric speeds, ie. transmit at one speed and receive at a different speed. Line conditions can and do vary in each direction. If your modem is capable of this it may connect at 28800 transmit and say 24000 receive. However if it isn't BOTH transmit and receive will use 24000 !
Cheap modems will also result in more disconnects whereas a good modem may be able to handle changes the line conditions gracefully and continue. They will also bump up the speed when the line condition improves. Cheap modems on the other hand will drop their speed lower and lower until they finally give up and disconnect .... in the middle of your 3 Meg download !
In general when it comes to modems (and most other things), You Get What You Pay For.

ERROR CORRECTION

Modern modems, that is V32bis and newer, have an Error Correction (EC) capability, that is they can detect and correct small errors in the transmission on the fly, saving your software from having to retransmit the corrupted data. This obviously improves download speeds. Some external modems have some indication to show if Error Correction is turned on (both modems on each end must have error correction turned on). On my Courier, it is the ARQ light.
If line conditions deteriotate to the point where the EC cannot handle it, the modems will turn it off and leave the software to retransmit the corrupted data.
Some modems may be set up to disconnect if they cannot negotiate or maintain an error corrected link. Did you know that ? Check your modem manual to see if this is enabled by default. This is fine if you absolutely want maximum performance and want to dial until you get a good link. However, imagine what happens when you dial, get a good link, EC is on, start a BIG download. Then midway, or at 90% completion as it usually the case, line conditions deteriorate, EC is lost, and your modem drops the connection. Nasty thought !
On top of that did you know Win95 dial-up networking has settings that do this ? Check out the Advanced settings, Error Control...there is a check box for "required to connect"...ie You MUST have an error corrected link to even stay connected. Modems will give up trying to correct for errors if the bit error rate gets too high. If your modem/Win95 is set up to do so, it will disconnect rather than continue, even though line conditions may be transient. Those of you lucky/rich (I'm in the lucky category) enough to own a US Robotics Courier V.Everything, may notice the ARQ light go off sometimes and comes back...the ARQ light indicates the error correcting status.
I was online one evening then an electical storm was approaching. I could see the lightning in the distance and hear the thunder. Suddenly I noticed my ARQ light on my courier blinking, data transmission stopped for a few seconds, the modem recovered, the ARQ light came back on and the data transfer resumed. What happened was a nearby (1-2 KM away) lightning strike induced spikes in my connection and caused errors. The errors exceeded the ability of the modem to correct for, the modem abandons attempting to correct for errors and the ARQ light drops. When the interference was gone, the modem renegotiated the error correction and resumed normal operation. I managed to complete my data transfer and disconnect before the storm arrived. Try that with a cheap modem.

CHECKING OUT YOUR PHONE LINE

A word about phone lines. Try this, pick up your phone, listen, and don't dial anything...hear that dial tone ? Ok that is getting in the way for you to listen to the LINE. Now press ONE number, dial tone goes away right ? Now LISTEN to the LINE. What do you hear ?
Crackling (sounds like crumpling paper or foil, or eggs frying !) ? Bad connection somewhere, check your internal wiring or call your phone company to fix it.
Crossed line (hmm, who's that talking on your line) ? 'nuff said, stop eavesdropping and call your phone company to fix it.
50/60 Hz hum ? your phone line is too close to some mains power, relocate it.
You should hear nothing except a very faint hiss, like air leaking from a tyre ... this is line noise and is normal, all electrical/electronic systems have a background noise.
Ok, so your phone line SOUNDS perfect but you STILL get 24000 or less when you connect. Why ?
1. If you have a "low cost modem", see above. 2. You house/office may be a long way from an exchange (switch). In general if the distance exceeds 4 KM from an exchange you may face significant signal loss.
3. Your phone company installed a DPG device on your line.

Thats it for now. I'll keep adding stuff as I find out more about how modems work.


This page was created on Wednesday, December 10, 1997 and last updated on Wednesday, December 10, 1997