On Sat, Apr 17, 2010 at 3:17 PM, hteller <[email protected]> wrote:
> Brian, > > > > I often hear people talking about preserving bandwidth and I wonder why > that is such a goal? Sometimes increasing bandwidth can improve performance > to the point where you get all the bits through in a shorter period of time, > making the channel available to others. Think about spread-spectrum or > packet radio vs. RTTY for example. > > This the argument that Winlink tried to foist upon everyone over five years > ago when asking for unlimited space. The fallacy of that argument is that > the frequency (or channel) is dominated by one station that keeps the > channel occupied until "its" traffic is passed, whereas, if more stations, > say five, can fit in one of our limited frequency ranges and pass traffic > simultaneously at only 1/5 the rate, the chances of more than one station > successfully passing traffic, even at a slower rate, is greater than the > chances of a single wide station passing its traffic. In the limit case, the > wide station keeps trying to pass traffic without any success at all, but > prevents other stations from using the frequency when one or more of them > could have been successful in passing traffic. > Welcome to the wonderful world of media access control or MAC. You have just jumped into one of my areas of expertise. MAC is deciding who gets to use a shared medium now. Packet radio actually made a [bad] attempt to deal with this and used straight carrier-sense multiple access (CSMA). The problem with that is, if you have two stations that have traffic, both transmit. Phil Karn pointed out the problem and we got p-persistent CSMA which made things much better but since it was never automated (persistence 'p' should be set to 1/(n-1) where 'n' is the number of stations with traffic to pass so 'n' is always changing) it never made much difference. In the case of HF the right answer is most likely a combination of routing, slot, and spectrum negotiation. The software controlling the stations should flood their time-bandwidth requirement to the other stations so that they can agree who will transmit where and when. The reason for flooding is that not all stations can hear all other stations so some stations will have to relay the information to other stations. (This problem gets much simpler with a single, centralized controller -- net control station.) Unattended PACTOR is not bad because of its bandwidth, it is bad because it doesn't implement any sort of MAC and it doesn't do any sort of CSMA to ensure that it doesn't interfere. As a modem it is the only adaptive game in town. As a protocol it is an interesting case study. As a MAC, it sucks. We can do a LOT better. As path loss decreases, it makes sense to increase bandwidth in order to reduce transmission time. This is a good lesson to learn from PACTOR. The optimal bandwidth is going to be a function of path loss. As path loss increases, you want to decrease bandwidth in order to keep power, and hence Eb/No, where you need for reasonable bit error rates. As path loss decreases you want to increase bandwidth bits/symbol, and signaling rate. > > The same it true for spread spectrum, or even HSMM. The idea is that a > station is on for only a short time, so it creates QRM on that frequency for > only a short time, and that is supposed to be acceptable for the harmed > station, as they can just repeat their transmission, or perhaps the signal > is so weak it is not noticed. The problem occurs when many spread spectrum > stations try to use a band of frequencies at one time. In the aggregate, > they will QRM the majority of stations over the band in question, since > there are few places where others can operate without any QRM at all. > This actually gets into coding robustness. It is not nearly as simple as you suggest. We can build codes that expect QRM. For example, some CDMA systems do 'puncturing' (actually not sending some bits) to do power control. This works because the error-correcting-codes fix it in the receiver. > In layman's terms, a single wide signal can blanket the entire passband of > a receiver (let's say MT63-2000), > What's wide about that? It is only 2kHz. That is nowhere near the entire passband of a reasonable receiver and even less a part of the entire band. My receiver has a passband of about 160kHz without retuning the LO. > and it is not possible to filter out that signal. > Sure it is. It is quite easy to filter that out. You have to wait until it is finished passing traffic, no matter how fast it > transmits. > Part of the problem here Skip is that you are thinking in terms of how things are instead of how they should be. Think about it for a minute. Do you think that the cell phone system hasn't already solved the MAC problems? The cell phone system is a automated, radio-based, multiple-access, traffic-passing system. So such a thing is possible. HF does offer us some interesting challenges but that doesn't mean we have to keep doing things the old way. > In the extreme, if you use a 50 Hz wide IF filter for PSK31, 40 stations > can occupy the IF bandwidth at one time without any of them capturing the > AGC or interfering with each other. > Yes, and the codec should be selecting the receiver characteristics depending on the modem and the coding. > Again, signaling theory can probably make a scientific argument for > conserving bandwidth so a band of frequencies is available to more than one > station at any given time. > Definitely. > > All the "get on and get off fast" arguments may be valid on wired circuits, > where access is controlled, but they tend to fall apart in practice on the > HF bands where space is so limited and access is random. > No one has done them! I don't think I have ever heard of an intelligent MAC on HF. I would really like to see one and if one exists, I would like a pointer so I can study it. I have worked on RF-based data distribution systems where bandwidth is dynamically allocated to stations based on traffic need using both TDMA and FDMA. You throw CSMA in there on top of it to deal with rogue stations that are not part of your system. > As a recent example, ROS uses spread spectrum, occupies 2200 Hz of > spectrum, but currently is only able to handle two stations on the frequency > at one time without interference between stations, and the weak signal > performance is no better than MFSK16, for example, in 1/6 the bandwidth. > So? That means that ROS is not very good. OTOH, what is it throughput in bps? > The simple reason that narrow bandwidth is such a goal is that it gives > more stations the "opportunity" to use a limited number of frequencies at > the same time, and speed of traffic passing is not so important for a 50 wpm > QSO, which is the speed that G3PLX chose as a reasonable speed to > accommodate a most typists. PSK63 is twice as wide, so only half as many > stations can use the same frequency range, but the speed (100 wpm) is > greater than needed for ragchews. It was only intended to be a RTTY > replacement for contests, in less bandwidth, but just as fast as RTTY > overall. However, there is a penalty of 3 dB in minimum S/N for the extra > speed, so for ragchews, PSK31 is sufficient and twice as many stations can > use the same range of frequencies simultaneously. > But when the path provides less loss, you can take advantage of that extra S:N to do one of two things: increase speed (less energy needed per bit) or decrease power. And who cares how fast someone types. Why should the transmitter be on for the entire time the person is typing? Why not transmit when some or all of the message is ready? Think packet radio. It doesn't transmit until you have finished typing a line (or a whole message for that matter). Why such a marriage to having the transmitter on for the entire time the person is sitting there thinking or typing (badly for most of us ;-)? Why the marriage to a 3kHz bandwidth? Why keep doing things the way we have been doing things? Why Why Why? We now have massive amounts of compute power in virtually every shack. We have SDRs that provide hundreds of kHz of bandwidth. They also provide infinitely-variable, brick-wall filters with zero passband ripple and constant group delay. Radios are frequency agile. There is no reason that the computers can't build up a map of what part of the band is being used and what isn't, then automatically tune to the empty spaces. Throw in a bit of communications on the side (signaling) and radios can relay which frequencies are in use so there won't be any QRM. And I am sure someone will point out that packet radio doesn't work well on HF. I know. But that is because it uses a lousy modem and is badly designed. That doesn't mean that some other form of packet radio with a proper adaptive modem and an intelligent MAC won't do a MUCH better job. > That is how I see it from a layman's viewpoint... > I see it from a different viewpoint. -- 73 de Brian, WB6RQN/J79BPL
_______________________________________________ fldigi-alpha mailing list [email protected] https://lists.berlios.de/mailman/listinfo/fldigi-alpha
