EBU Technical Review

No. 305 (January 2006)

Spectrum efficiency

As the radio spectrum is a valuable resource, it is obvious that all users should use the spectrum efficiently. But how do you measure spectrum efficiency?

If you ask an engineer, you will probably be told that spectrum efficiency is measured in bits/second/Hz – thus quantifying the data rate that can be transmitted (and received) within a given bandwidth. Looking at the performance of practical digital broadcasting systems, we find that digital terrestrial TV transmissions using DVB-T can deliver data rates of 24 Mbit/s in an 8 MHz channel – corresponding to a spectrum efficiency of 3 bit/s/Hz.

Is this good, bad or indifferent?

Fig. 1 Data rates offered by DVB-T transmissions in an 8 MHz channel as a function of carrier-to-noise (C/N) ratio (assuming a channel with Gaussian noise)

Fig. 1 shows the performance of various options permitted by the DVB-T standard (ETSI EN 300 744). Note that the higher-order modulation schemes, such as 64-QAM, perform much better than QPSK – at least in terms of data throughput in an 8 MHz channel with Gaussian noise. However, Fig. 1 also shows that this improved performance of 64-QAM requires a significantly higher carrier-to-noise (C/N) ratio – more than 10 dB higher.

To me, the most important message conveyed by Fig. 1 is that there is no point in quoting a single value for spectrum efficiency (as measured in bits/s/Hz) simply because spectrum efficiency is dependent on the C/N ratio. As the C/N ratio is increased, much higher data throughput becomes possible. Hence, when judging the relative performance of systems, you should quote spectrum efficiency at a given value of C/N.

The solid curve in Fig. 1 shows the Shannon limit, which represents the absolute limit in this respect. The DVB-T standard is, at least, 5 dB worse than the Shannon limit – but the performance of the DVB-S2 standard (ETSI EN 302 307) approaches within 2 dB of the Shannon limit.

When designing communications systems, spectrum efficiency and C/N ratio cannot be considered in isolation. It can be argued that they are equally important. Increasing the received C/N ratio by 10 dB would dramatically improve spectrum efficiency: this could be achieved by increasing the transmitter power by 10 dB – equivalent to increasing power by a factor of 10. In practice, such an increase in power is unlikely to be feasible because of economic reasons and because other users of the spectrum might not welcome a 10 dB increase in interference!

Instead of increasing the transmitter power by 10 dB, you could demand the use of high-gain receiving antennas as in the case of many DVB-T services where broadcasters assume that viewers will install roof-top antennas and, hence, the service is designed to deliver, say, 24 Mbit/s (equivalent to 3 bit/s/Hz). On the other hand, in the case of DVB-H services intended for reception on handheld devices, it is necessary to operate at much lower values of C/N ratio and, consequently, at lower data rates — between, for example, 3.75 Mbit/s (equivalent to less than 0.5 bit/s/Hz) and 11 Mbit/s (equivalent to 1.375 bit/s/Hz). If spectrum efficiency were the only criterion for selecting transmission parameters, everybody would choose DVB-T at 24 Mbit/s or more – but spectrum efficiency must be sacrificed if network operators (and consumers) want reliable reception on handheld devices.

So where does this leave the concept of spectrum efficiency? First, we must acknowledge that measuring spectrum efficiency in bit/s/Hz is inadequate and inappropriate. Secondly, we need to extend the definition of spectrum efficiency to take account of other factors, such as transmitter powers, the overall cost of networks and, crucially, the potential for interference to (or from) other services.

Instead of trying to maximise spectrum efficiency, you could go in the opposite direction by designing a system to operate with very low values of C/N ratio such as -10 dB (implying that the noise is 10 dB stronger than the wanted signal). In such circumstances, the Shannon limit is 0.14 bit/s/Hz. At first sight, this seems so low to be of no interest, but it corresponds to a data rate of 1.1 Mbit/s in an 8 MHz channel. Although it is difficult to approach the Shannon limit, practical systems based on very wide bandwidths can deliver significant data rates despite being “below the noise level”.

In the past, wide-band techniques – such as spread-spectrum – were mainly limited to military applications. However, in 2002, the USA’s Federal Communications Commission (FCC) licensed Ultra Wide Band (UWB) services operating in the 3.1-10.6 GHz band. Commercial applications of UWB technology, such as in-home networks for distribution of video and audio, offer data rates of 100 Mbit/s over distances of up to 10 m.

It is frequently claimed that such devices will not cause any interference to other services because they operate “below the noise floor”. Despite these seductive arguments, the reality is that such systems will increase the noise floor. Actually, it is not sensible to talk about the noise floor as if it were uniformly flat. Distant sources of noise and interference do tend to produce a relatively constant noise level, but a UWB device operating at a distance of, say, 1 m from your receiving antenna will dramatically increase the noise floor – more like a noise mountain!

In the analogue world, it did not make much difference if the noise floor increased by 1 dB – because the S/N ratio might change from 30 dB to 29 dB. In the digital world, there is a margin of only a few dB between perfect reception and no reception at all. To make matters worse, most listeners and viewers unknowingly operate their digital receivers very close to the failure point. In these circumstances, a small increase in the noise or interference levels can cause complete failure of digital reception.

UWB and similar systems will be unlicensed or licence-exempt (as with wireless burglar alarms or Wi-Fi systems). A key problem with unlicensed devices is that, even in the event of serious interference, there is no licence to be revoked! The users of the spectrum are simply expected to resolve interference problems. What will happen if your TV or radio reception suffers interference from an unlicensed system installed by your neighbour? Unless you can identify the source of the interference, you will be irritated to find that your digital reception has failed – or, at best, fails occasionally. Unfortunately, interference to digital broadcasting systems is incredibly difficult to diagnose – even if you are a highly-knowledgeable engineer.

At present, there does not seem to be any plans to introduce such systems into the bands allocated for broadcasting – but this is almost certainly a matter of time! However, broadcasting bands have already began to suffer interference from other unlicensed systems – often, in fact, from systems that do not intentionally radiate, such as ADSL services on twisted-pair telephone lines and systems for delivering broadband services over electrical networks (known variously as BPL, PLT or PLC). Experience to date suggests that properly-installed ADSL services are relatively benign, but the power line technologies are a major problem for reception of AM radio services. One regulator personally told me that the objective of extending the use of broadband access to the Internet was politically far more important than the need to protect old-fashioned AM radio services (especially in the HF bands) – but he was surprised to learn that Digital Radio Mondiale has spent a huge amount of time and money developing a new digital radio system to operate in these bands. If power line technologies become widespread, they will radiate noise across the HF bands and jeopardise the success of Digital Radio Mondiale.

When you hear somebody telling you that their system offers unrivalled spectrum efficiency and/or does not cause any interference, remember that you are rightly suspicious of used car salesmen!

Philip Laven
Director
EBU Technical Department

23 January 2006

Editorials Archive