Slashdot Mirror
CSIRO Demonstrates Fastest Wireless Link Yet
rob101 writes "The CSIRO yesterday demonstrated the world's 'fastest' wireless radio link by transmitting sixteen full quality DVD streams over a 250m link and only using a quarter of the available bandwidth. 'The CSIRO ICT Centre today announced that it has achieved over six gigabits per second over a point to point wireless connection with the highest efficiency (2.4bits/s/Hz) ever achieved for such a system.'" CSIRO hopes to double the speed of this connection in the future, pushing twelve gigabits a second.
"Don't lose your mind trying to set it free..."
Australia's network covers huge areas with a spare population, it uses radio and/or sattelite links to link remote exchanges to the trunk. During the late 90's I had extensive experience with an Australian wide mobile application, back then the radio links had a 2500 baud connection. Arguing about service to the bush is a political constant that hasn't changed in the last few decades.
And did you exchange a walk on part in the war for a lead role in a cage? - Pink Floyd.
CSIRO is Australian.
Your country does indeed take this sort of technology, and doesn't like to honour the patents on it either! So stop complaining.
Actually, the reason you won't have that at home is that it operates at 85 GHz, so it's only useful for line of sight communication. Think next-generation microwave towers, not WLAN.
Well, the result I found was 20 TB for the books (163840 gigabits) and they said in the article it was 6Gb/sec, so 6/163840=0.00003662109375 LOC/sec.
"Don't lose your mind trying to set it free..."
A "hertz" is a cycle per second, so what they're really saying is that they're getting 2.4 bits per cycle of the carrier. I agree that bits/sec/Hz is a ridiculous term for someone to have made up, but they would probably justify it by pointing out that there's not just one effective "carrier" in a modern modulation scheme.
In any event 2.4 bits/s/Hz is nothing special, unless it applies to individual subcarriers in an OFDM-like scheme. 802.11g sends 54 MB/sec in a channel about 20 MHz wide, for a bandwidth efficiency of 2.7 bits/second/Hz. Sounds like they basically threw a metric assload of RF bandwidth at the problem (another technical term found in relatively-few EE textbooks).
Dahlmann tightly grips the knife, which he may have no idea how to use, and steps out into the plain.
CSIRO == Commonwealth Scientific and Industrial Research Organisation,.. and it does really good work, and provides more than the average bang for the buck. If only US companies would pay for using patented technologies the CSIRO develops Australia would have an even greater return on investment, or even more practical research to aid industry in Australia, and in turn the world.
No, not the frequency of the channel - the width of the channel.
Look up the Shannon-Hartley theorem on Wikipedia for some context. It establishes the theoretical maximum capacity given the signal to noise ration and the width of the channel.
That's bits per second per Hertz of BANDWIDTH, not transmission frequency. Or at least it better be, because that's the only thing which makes any sense to consider (the carrier frequency has no bearing on the data rate).
I might be wrong, radio comms stuff is not a strong area of mine, but I thought the intention of numbers such as these was to provide a view of the efficiency of the system. For example a QAM system gets 4 bits per cycle, 64-QAM gets, well, 64 (I think it's used in DVB broadcasts in the UK, don't know about elsewhere).
As the amount of information per cycle goes up, the risk of noise and corruption increases, since they have a more significant effect on the signal. Seeing as there is usually a fixed frequency band to work with, then it is necessary to increase the number of bits/cycle in order to increase the throughput.
It sort of reminds me of the clocks per instruction measure for cpus.
God, I wish there were a -5 "Totally Wrong" moderation.
Carrier frequency has nothing to do with how much information a channel can carry. Channel bandwidth (spectrum used on each side of the carrier frequency) is what matters.
For example, a 6 MHz channel at 450 MHz and one at around 800 MHz have the exact same channel capacity (assuming that the SNR at the receiver is the same on each channel.)
To be specific, the formula for maximum channel capacity of a communications channel is given by Shannon's Law:
C = W log (1 + Eb/No), where Eb/No is the signal to noise ratio of the channel and W is the channel bandwidth.
Maximum C for a given SNR and W (or minimum SNR for a given C and W) is not achievable in practice, but recent advances in error control coding techniques such as LDPC and turbo codes have allowed people to get to within just 1 dB of the minimum SNR for a few years. (And yes, this technology is in cell phones. If I recall correctly, turbo codes are used on some cell phone downlinks when transmitting image data that is not latency-sensitive. Unfortunately both turbo codes and LDPC both introduce pretty high latency to a communications system.
2.4 bits/sec/Hz is nothing new. As others have pointed out, plenty of other systems have been doing this for quite some time.
Cable modems - I believe the DOCSIS maximum limit is 36 Mbits/sec over a 6 MHz channel. 6 bits/sec/Hz - the nice thing about cable distribution is that the inverse square law goes bye bye and high SNRs are easily achievable.
ATSC digital television - 8VSB provides 19.2 mbits/sec over a 6 MHz channel. Just over 3 bits/sec/Hz over relatively long free-space distances, although transmitter power is measured in kilowatts.
There isn't really enough information to figure out exactly what they did, but it looks like the CSIRO people just threw a massive amount of channel bandwidth at the problem. 2.4 bits/sec/Hz means their SNR was not that high.
BTW, yes, it IS true that at higher carrier frequencies, there is more free spectrum available to use wider channels, but there is no direct link between carrier frequency and channel capacity as you claim.
retrorocket.o not found, launch anyway?
"Doesn't the carrier freq needs to be > 2 times the data per Shannon?"
No, that's Nyquist sampling. To sample an analog signal without aliasing, the sampling rate needs to be 2x the bandwidth of the input signal. Doesn't directly apply here, although it does govern how fast a receiver ADC must be for a software defined radio. NOTE: Carrier frequency does not impose any requirements on the ADC, only channel bandwidth. i.e. an ATSC digital television signal needs at least a 12 MHz sampling rate to be properly sampled, as it is approximately 6 MHz wide regardless of channel carrier frequency.
Shannon's Law states:
C = W log (1 + SNR)
C = channel capacity
W = channel bandwidth
SNR = signal to noise ratio of the channel
Thus, achieving 2.4 bits/sec/Hz is easy - just increase your transmit power or your channel gain to increase SNR. This is why cable modems easily achieve 6 bits/sec/Hz (DOCSIS upper limit is 36 Mbits/sec over a 6 MHz channel, any lower speed is an artificial cap from your provider) - when you are transmitting over a cable instead of free space, losses are (comparatively) low and hence high SNRs are not difficult to achieve.
In this case, it appears the CSIRO guys just threw a lot of bandwidth at the problem (large W).
Easier said than done in the real world. Fixed point-to-point links are easy (directional antennas reduce multipath significantly, what multipath does remain does not change rapidly so requires little receiver processing power to estimate and compensate for.) Mobile environments with rapidly changing high amounts of multipath are where the real challenges are, and thanks to Moore's Law, technology is growing by leaps and bounds in this regard. Error correction techniques known since the 1960s but not implementable until recently (such as LDPC) are now in regular use thanks to increased computing power.
retrorocket.o not found, launch anyway?
Not so fast, you're forgetting a logarithm (base 2) here: 1 bit gives 2 possible values, 2 bits give 4 distinct combinations etc...
So QAM, using 4 distinct phases of the carrier transmits 2 bits per cycle. 64-QAM requires you to distinguish between 64 different possibilities for each cycle, and that gets you 6 bits. Whether it is actually possible to distinguish the different possibilities depends on the signal-to-noise ratio.