GB2449858A

GB2449858A - Improved cyclic delay diversity

Info

Publication number: GB2449858A
Application number: GB0710666A
Authority: GB
Inventors: Oliver Haffenden
Original assignee: British Broadcasting Corp
Current assignee: British Broadcasting Corp
Priority date: 2007-06-04
Filing date: 2007-06-04
Publication date: 2008-12-10
Anticipated expiration: 2027-06-04
Also published as: GB0710666D0; GB2449858B

Abstract

A transmission system is disclosed that address flat fading problems found particularly in single frequency networks (SFNs). A spread of cyclic delays are applied to the signal at the second or further transmitter antennas to produce a number of cyclically delayed versions of the time-domain signal, such that portions of the signal are delayed by different amounts. As a result, the time synchronisation algorithms at the receiver are persuaded to respond to the centre of the delay spread, giving good Inter Symbol Interference performance. The apparent spread can be achieved by having the apparent delay vary as a function of frequency across the signal spectrum.

Description

Delay Diversity Transmission Apparatus and Method The invention relates

to a delay diversity transmission apparatus and method, and in particular to those suitable for use with technologies such as Digital Radio Mondiale (DRM) and Digital Video Broadcasting (DVB).

Flat fading describes the situation in which an entire radio frequency (RE) signal experiences deep attenuation, in contrast to frequency selective fading in which some parts of the frequency spectrum are faded and some are boosted. One way in which flat fading occurs is by simple attenuation: if the direct path from the transmitter to the receiver is blocked, for example by a large building. Another cause is destructive interference between the direct signal and an echo resulting from the signal reflecting off an object such as a building or other topographical feature In general, there will be a delay between the two paths taken by the signal because of the different path lengths As the phase difference arising from this delay will vary with frequency, the interference will be constructive at some frequencies and destructive at others. On a large scale, the result is selective fading, but as the pitch of the selectivity (the frequency difference between the successive nulls) is equal to the inverse of the relative delay, for a short enough delay the nulls can be so wide that the entire signal spectrum falls into a null and effectively suffers flat fading.

One way to mitigate this problem is to use a wide spectral occupancy, so that even a short delay gives rise to at least one cycle of fading across the bandwidth.

In this case, only part of the spectrum will fall into the null. Clearly, this will reduce the spectral efficiency unless the data rate is increased by way of compensation.

Thus, it is desirable to convey several different services in the wider bandwidth.

One option is to convey a large number of services in a single high bit-rate multiplex; this is the approach adopted by Digital Audio Broadcasting (DAB).

Other possibilities include spread-spectrum techniques such as Code Division Multiple Access (CDMA) and frequency-hopping whereby multiple transmissions share the same, broad channel, However, utilising the signal bandwidth in this way is not always possible.

DRM+ for example is a proposed extension of the DRM system to the VHF bands For various reasons it has been decided that the DRM+ system should be designed such that one DRM+ signal carries only one service, without need for co-operation between broadcasters sharing the same band. Furthermore, it is possible that DRM+ might be introduced gradually into VHF Band II where it would exist for some time in parallel with analogue FM transmissions. The need to squeeze DRM+ signals in between FM signals is likely to restrict the maximum RF bandwidth to around 100kHz.

The table in Figure 1 compares the nominal occupied bandwidths of several digital broadcasting systems. In each case, the table gives the minimum relative delay between two paths that produces at least one cycle of selective fading across the bandwidth This is also expressed as a difference between the two path lengths Referring to the table, the existing DRM system for use below 30MHz is seen to be the most vulnerable to flat fading. However, this is not generally a problem since multipath at high frequencies (HF) is normally the result of ionospheric propagation and tends to involve path differences of 1 ms or greater.

Nevertheless, if this system is used for local broadcasting in the 26MHz (urn) band, propagation is more VHF-like and flat fading can be a problem.

DAB and DVB-T (Digital Video Broadcasting -Terrestrial) have relatively w1e spectral occupancies and correspondingly short minimum delays. The US HDRadio system has two digital blocks, one on each side of the FM signal, and so occupies up to 400kHz in total, making it more resilient to flat fading than DRM+ However, it is still possible that a selective fade could have nulls falling on both of the digital blocks, so this column in the table should not be taken entirely on face value.

Flat fading will occur at points in space where the relative delay is less than the minimum value discussed above, and the relative phases are such that the interference is destructive rather than constructive. In general, moving the reception point will change the relative phases so that the interference eventually becomes constructive. How far it is necessary to move depends on the geometry of the reflection and of the direction of movement. Another way to look at this is to say that there is standing wave pattern in space, with alternate contours of constructive and destructive interference. In general, it will not be necessary to move by more than half a wavelength to escape from a deep null, although if the receiver is in a car which happens to be driving along a contour of flat fading, there will be no escape One way to address this is to use receiver spatial diversity, in which there are a number of receive antennas positioned on the order of half a wavelength apart.

This is increasingly common for car reception. At VHF band II, the wavelength is about 3m, so antennas on the corners of a car roof are likely to give good performance. However, this type of diversity is expensive, since it typically requires multiple receiver front-ends as well as multiple antennas, and is impractical for a portable or hand-held receiver whose physical size is too small to give sufficient separation of the antennas. For DRM below 30MHz, the wavelength rises to more than lOm, whereupon only a large truck will give the necessary separation For these reasons, we have appreciated that it is desirable to have only a single receive antenna, but to transmit from two different antennas configured in a transmitter diversity scheme. In principle, if the signal from one antenna is faded, the other antenna is arranged so that the signal is likely to experience constructive interference because of the different antenna geometry.

Delay diversity is one such scheme, in which two antennae are fed the same signal, but with a delay of T seconds between them. Provided the delay seen at the receiver is greater than the reciprocal of the occupied bandwidth, there will be at least one cycle of selective fading across the signal. One problem with the scheme is that it reduces the effective guard interval, because in most locations some signal will be received from both transmitters meaning that there will already be a delay spread of T. Any naturally arising multipath will add to this.

Another drawback is that it introduces selective fading in cases where signals are received from both transmitters, using up some of the transmission system's capacity to correct errors, and consequently increasing the required signal to noise ratio (SNR).

Summary of the Invention

The invention is defined in the ndependerit claims to which reference should now be made. Advantageous features are set out in the dependent claims,

Brief Description of the Drawings

A preferred embodiment of the invention will now be described in more detail, by way of example, and with reference to the drawings in which: Figure 1 is a table providing a comparison of various systems' vulnerability to flat fading; Figure 2 is a schematic illustration of a diversity signal transmission scheme; Figure 3 is a schematic illustration of cyclic diversity; Figures 4, 5 and 6 illustrate different ways of applying the delay spread in the preferred embodiment of the invention; Figure 7 is a graph of a frequency dependent function illustrating how the delay might be applied; Figure 8 is a block diagram illustrating a first transmitter apparatus; Figure 9 is a graph showing calculated group delay profiles for varying numbers of Bessel terms; Figure 10 is a graph showing the frequency response magnitude in dB for various numbers of Bessel terms; Figure 11 is a graph showing the total received power relative to a single transmitter for a range of delays, where N=6; Figure 12 is a graph showing the variation of received power with delay and phase (linear group delay).

Figure 13 is a block diagram illustrating a second transmitter apparatus; and Figure 14 is a block diagram of a receiver apparatus.

Detailed Description of the Preferred Embodiments

The preferred technique uses a variation of cyclic delay diversity, and will now be explained in more detail, with reference to Figures 2 and 3.

Figure 2 schematically illustrates the configuration of a transmitted signal in an Orthogonal Frequency Division Multiplexing scheme. Data is carried primarily in the active symbol parts of the signal, with respective symbols being separated from one another by a region called the guard interval. Two signals, slightly displaced from one another, are illustrated in Figure 2 as two versions of the signal are transmitted with a time delay between them. The time delay is either deliberate, as in a delay diversity scheme, or may arise naturally from the different effective path lengths from the antenna to the receiver.

A guard interval and an active symbol together form a symbol. To form the guard interval, information is copied from the end of the active symbol and placed at the beginning. This has the effect of preserving the orthogonality of the different frequency carriers used by the system, as well as maintaining a good SNR. This can be understood by considering what happens at the receiver for different delay lengths. If the receiver receives two copies of the signal, but delayed in time by more than a symbol duration, then the two symbols received represent different data and when the receiver attempts to interpret them they will interfere with each other as noise. If on the other hand, the delay is less than a symbol duration, then the delayed signal comprises some data from the current symbol, and some from the adjacent symbol. The data from the adjacent symbol would again be seen as noise, but the data from the same symbol can be processed constructively by the receiver boosting the strength of the received signal. The guard interval is a technique for extending the effective width of the active symbol, by copying some of the information from the end of the active symbol to the beginning. In this way, the second signal can be delayed by as much as the duration of the guard interval before data from the adjacent symbol forms interference. This situation is shown in Figure 2, in which two copies of the signal are displaced by the maximum delay that can be tolerated. The range of samples processed by the receiver (the receiver's FFT window) is illustrated by the dotted lines As noted, above the drawback with delay diversity schemes, is that the sum of the deliberate delay introduced by the transmission system and the delay resulting from propagation should not be longer than the guard interval. Thus, the width of the guard interval is effectively reduced.

Cyclic delay diversity is a technique that provides the same benefits as delay diversity without reducing the effective guard interval. Instead of introducing a true delay, the samples making up the active symbol are cyclically rotated before the guard interval is added, as shown in Figure 3. It should be remembered that the data in the active symbol is periodic, which is what allows the guard interval to be added to the beginning of the symbol in the first place Thus, cyclically rotating the data in the active symbol does not change the data, only its phase.

Provided the receiver's FFT window is aligned with the active symbol part for the earliest arriving path, cyclic delay diversity systems can tolerate real-world delays up to the guard interval duration without any inter-symbol interference (ISI) or loss of orthogonality resulting in inter-carrier interference (Id). Once again, a signal that has experienced this maximum delay is shown in the figure to illustrate this Note that the signal from the transmitter that does not have the cyclic delay will have the guard interval and active symbols exactly aligned in time with the cyclically delayed version, so the same range of delays can be tolerated from both.

Although the cychc delay has not changed the symbol timing, the signal seen in the receiver's FFT window will appear to have been delayed, just as though simple delay diversity had been used. Remember that the magic of the guard interval is that it turns a real delay into a cyclic delay. The pattern of constructive and destructive interference, i.e. the selective fading resulting from the combination of the signals from the two transmitters will be the same as in the simple delay case.

Looked at another way, the cyclic delay means that the frequency domain signal has been multiplied by a complex exponential e*I, so the carriers have had their phases rotated by a spiral: hence the alternative name "phase diversity". This continuous change of phase across the spectrum is what leads to alternate constructive and destructive interference when added to the constant phase of the non-delayed signal. Consequently, the considerations as to where and whether flat fading will occur are the same as in the simple delay case.

We can see that, in principle, the introduction of the cyclic delay has achieved the same result as the simple delay, but without using up any of the guard interval, provided that the receiver places its window in the position indicated. Whether this happens in practice will depend on the time synchronisation algorithm.

There are two main approaches to time synchronisation in OFDM. pre-and post-FF1 techniques. Pre-FFT techniques are based on "guard-interval correlation", in which the signal is multiplied by its conjugate across a delay equal to the active symbol period. The basic principle is that there is correlation when the two samples being multiplied come respectively from the guard interval of a symbol and the part at the end of the symbol from which the guard interval is copied. If we consider applying this operation to the cyclically delayed signal, it is easy to see that the correlation will occur at the correct time, i.e. the guard interval will be correctly identified. In fact, the cyclically delayed signal is just an OFDM symbol of the original timing but with different modulation (because of the action of the complex exponential), and the guard-interval correlation technique does not take into account the modulation.

Post-FFT techniques generally derive an estimate of the impulse response based on the frequency response, which is in turn derived from the pilots. Assuming that the cyclic delay has been added as described, the phases of the pilots will be exactly as they would have been for a real delay, so the estimated impulse response will be shifted later in time. Taking the time domain view, it is impossible to tell the difference between a real and cyclic delay given only the signal from the receiver's window, which forms the input to the FFT. The figure shows the apparent timing of the signal which has been delayed by Tg in addition to its cyclic delay The synchronisation algorithm will think that this is too late for the current position, and is likely to adjust the timing to try to find a compromise between the early path with no cyclic delay and the late path with its cyclic delay.

To avoid this, the true delay must be reduced by the amount of the cyclic delay.

It is therefore likely that post-FFT time synchronisation techniques will incorrectly position the receiver window as though the cyclic delay were a real delay.

Unfortunately, post-FFT techniques are generally preferred as they give better performance, pre-FFT methods are usually used only to achieve the initial coarse synchronisation.

So far, we have assumed that the cyclic delay had been added in the time domain as shown in Figure 2. Since it corresponds to multiplication of the frequency-domain signal by a complex exponential, it could however equally be carried out in the frequency domain This gives us the option of applying the cyclic delay only to some carriers and not to others. If we chose not to apply the rotation to the pilots, the resulting impulse response would not show the apparent delay and the synchronisation would be correct. However, the pilots are also used for channel equalisation, and this would now be incorrect. To overcome this, we could have two sets of pilots, one with the rotation and one without. This would increase the overhead and reduce the data rate, although it is possible that the synchronisation could make do with a smaller number of pilots.

Unless, some extra steps are taken, the post-FFT synchronisation is likely to adjust the timing as though there were a real delay. Although the technique apparently does not reduce the available guard interval, in effect it has reduced it by the same amount as a real delay because of the action of the time synchronisation. The fundamental problem with cyclic delay diversity is that there appears to be a real delay from the point of view of the post-FFT time synchronisation, and this algorithm is likely to try to compensate for it.

The preferred embodiment of the invention therefore makes the signal appear to have a spread of delays, equally disposed about the true delay value. In this way, the synchronisation algorithm is persuaded to respond to the centre of the delay spread, giving the best ISI performance. This apparent spread of delays can be achieved by having the apparent delay vary as a function of frequency across the spectrum. A constant delay corresponds to a phase shift that varies linearly with frequency (linear phase), so in the preferred embodiment the rate of change of phase with frequency is not constant. In other words, a deliberate group delay variation is introduced. This will be shown to give an extra benefit: the worst-case flat fading resulting from interference between the two transmitters can be reduced, which is potentially useful for more traditional Single Frequency Networks or SFNs The preferred embodiment therefore varies the cyclic or phase delay applied to carrier signals in the signal spectrum according to the carrier frequency. An apparatus for doing this will be described later. First, however, it is useful to consider the effect of the spread at the receiver.

If our apparent delay spread is symmetrical about the true delay, and only one signal is being received, we can assume that the signal is placed symmetrically in the middle of the safe range calculated by the receiver's synchronisation algorithm, as shown in Figure 4. This is generally how DRM receivers have been found to behave. This is also the desired behaviour, since it corresponds to placing the real delay in the middle of the lSl-free range.

Suppose now that a second signal is transmitted, without any kind of cyclic delay or apparent spread. Provided the total apparent extent of the delay spread is less than the guard interval duration, it will be possible for the receiver to fit all of the paths inside the lSl-free region. Indeed a good time synchronisation algorithm would be expected to do so. However, if the total delay spread arising from the deliberate delays from the transmitters plus natural multipath is close to the guard interval duration, the total delay spread seen by the receiver might exceed this limit, when the deliberate spread of the signal from one transmitter is taken into account. Figure 5 shows a simple SEN case; It can be seen that the time synchronisation algorithm of the receiver may need to find a compromise position and this might involve pushing the signal from the second transmitter out of the safe region. The real optimum in this case would position the two true delays at the beginning and end of the lSl-free region respectively.

In order to make this result more likely it has been found beneficial to apply the deliberate spreading at both transmitters, to produce the signal spread shown in Figure 6. in this case the receiver time synchronisation algorithm is likely to position the two signals symmetrically, which is the optimum position and leads to no reduction of the effective guard interval.

In the preferred embodiment the phase of the frequency response across the spectrum is varied so as to give a deliberate apparent delay spread. The receiver will only have a sub-sampled view of the frequency response because of the pilot pattern, and will need to interpolate between these sampling points. An important question is what we can do to the phase response without significantly upsetting the frequency interpolation.

It is not unreasonable to assume that the interpolator has been designed to deal with a channel that can be represented by an impulse response, in which the extent of the impulse response is limited to a range of delays. To accommodate the situation shown in figure 6, the frequency sampling corresponds to a range of delays that can be equalised (the delay timewidth) and the "equaliser timewidth" will need to exceed the guard interval duration; this is generally possible because the pilot spacing is chosen to give a Nyquist range that exceeds the guard interval duration by a significant margin One way to make an apparent delay spread that is disposed symmetrically around the true delay, is to start with a positive delay at one end of the signal spectrum and gradually move to a negative delay of equal magnitude at the other, with zero delay at the centre. Thus, in one example, the frequency dependent function for the delay or the frequency response' begins at one end of the spectrum as an anticlockwise spiral having a pitch that gradually opens out, changing direction in the middle, and leading to a clockwise spiral at the other end.

If we make the group delay vary linearly, we have a frequency response 11(f) = where the group delay d(f) varies over the range r1 across the occupied bandwidth B. d(f)=__L'= 2f 2.ir df rnafi so that the phase law is given by = We see that this results in a square-law for the phase, since the group delay is the rate of change of phase with (angular) frequency. Figure 7 illustrates the principle. An example transmitter according to the invention may therefore apply the delay spreading to the frequency spectrum, by multiplying each carrier of frequency f by the complex number given by this expression of H(f) and 43(f).

Apparatus for doing this is illustrated later.

In practice it would not be necessary to introduce such a large range of group delays and there might only be one turn of the spiral on each side of the centre.

However, the example is easier to understand initially when presented in this way.

The waveform shown in Figure 7 is a chirp, and it is useful to consider how well a receiver is able to deal with it We have assumed that the receiver's frequency interpolator is designed to deal with an impulse response that is time-limited to some range slightly exceeding the guard interval duration. To decide how well this frequency response meets this criterion we need to know the time-domain equivalent: its Fourier Transform There is an immediate problem: the ideal chirp waveform goes on for ever in frequency, with ever higher rate of change of phase. it also happens to be its own Fourier transform, so the corresponding impulse response is also infinite in extent and has constant amplitude: it is not time-limited as required. This is not surprising, since the spiral gets ever tighter corresponding to ever greater time delays.

Now, a DRM signal for example does not exist for all frequencies, as a result of which the parts of the chirp with large group delay will not be seen. Instead, the chirp is cut off at the edge of the occupied bandwidth by the receiver. To model this we could simply truncate the chirp, i.e. apply a rectangular window, and find the transform. However, this will not correctly represent the behaviour of the interpolator, because it assumes that the same interpolator taps are used right up to the edge of the spectrum even though there are no pilots beyond the edge to provide input to some of the taps: i.e. the taps start to fall off the end of the spectrum. In practice in DRM it is essential to deal with the edge carriers specially, by designing a set of progressively asymmetrical interpolators.

An alternative way of analysing it would be to apply some different type of window (e.g. Hamming, Hanning, etc), but again this would not represent the reality.

Furthermore, t would give less weight to the edges of the spectrum where the group delay is the greatest, so under-estimating the equivalent delay spread -a particular problem given that it is at the edges where the less well-performing one-sided interpolators are used. In practice we have found it reasonable to consider the equivalent delay spread to be equal to the range of group delay attained by the chirp across the spectrum.

The problem in the case of the chirp arose because it was a segment of an infinitely long signal with undesirable properties: specifically its infinitely long time-domain equivalent.

Instead of the chirp shown in Figure 7, an alternative signal is proposed having more acceptable time-domain properties, and having at least one segment which has the desired variable delay behaviour. As a general point, the signal should have constant amplitude because a rectangular spectrum gives the best frequency diversity. This allows us only to vary the phase 0(f).

With a chirp mentioned above, the group delay varies linearly so its logical continuation leads to ever higher group delays and an infinite delay spread in the corresponding impulse response An alternative is to have the group delay vary in a sinusoidal manner, as follows.

d(f) = 1 d (f) 1i Illax I 2ir df By choosing the correct portion of the sinusoid, an almost linear variation in group delay can then be achieved.

Since the group delay is the rate of change of phase with (angular) frequency, the phase itself will also be sinusoidal: 0(f) = rf0 cosJ The corresponding impulse response can be shown to be analogous to the standard case of phase modulation by a single large-amplitude sinewave Since the frequency response H(f) is periodic over f0 in frequency, it can be expressed as a Fourier series irrni0cos(iJ H(f)=e = H,,e where the H are the Fourier series coefficients given by Jo (2ij 2,'r 2 r in-i H=-Je e df f00 Letting + and x rmaxfo gives [I,, = 27r = (x) where J,, is the Bessel function of the first kind, order n. Used the standard result for the integral, and noting the integrand is periodic, any integration limits can be substituted encompassing a range of 2ii. The corresponding impulse response is therefore h(r) = jflJ (Tma fo)r --J This corresponds to a series of impulses in the time domain, i.e. a series of discrete paths. There are an infinite number of impulses over an infinite range of delays, and their amplitudes are given by the Bessel functions Jr,. However, as the Bessel functions fall off rapidly, the corresponding impulse response will be approximately finite in extent.

The response of the receiver can be improved by turning the analysis around.

Suppose a certain number of Bessel functions are chosen and a frequency response is synthesised from them:

N

H'(f) J".J,,(ninaxfo)e I') Because this is only part of an infinite series, the functions will not exactly reconstruct the constant amplitude, sinusoidal phase response desired. Instead the amplitude will not be exactly constant and the phase will not follow exactly the desired form. However, neither of these are serious limitations: COFDM can deal very well with a little passband ripple, and all that is necessary with the phase was group delay variation; the exact form of this variation is not important.

In the preferred embodiment, a finite number of cyclically delayed versions of the time-domain signal, or a finite number of constant-amplitude frequency responses are therefore superimposed on the original undelayed signal. This could be implemented in either domain, but the frequency domain method is preferable since the spacing between thedelays, equal to the reciprocal of f0, does not necessarily have to correspond to a whole number of time-domain samples This would require interpolation in the time domain, making it unnecessarily complicated.

Figure 8 is a block diagram showing an example apparatus 10 for implementing the invention in the frequency domain. Each block may be implemented in software or hardware, according to the design choice.

The apparatus is based on an OFDM transmitter, and so includes conventional OFDM transmitter stages, that will now be explained in a little more detail.

Analogue data is first received by source encoder 12 and converted into a bit stream of digital data. The bit stream is subsequently passed to channel encoder 14 where it is processed to be more suitable for transmission. In particular, the channel encoder may scramble the bit stream using a pseudo-random number generator to give the signal better time-domain properties when transmitted, as well as applying Forward Error Correcton in order to provide redundancy in the signal for resilience to errors. The Constellation Mapping section 16 receives the signal output from the Channel Encoder and maps it section by section of a predetermined number of bits into a complex number according to a predetermined encoding. The constellation is the name given to the plot of each complex number that is possible according to the encoding on an Argand diagram. Each complex number is then modulated onto one of a plurality of a predetermined carrier wave by the Carrier Mapping section 18. The output of the Carrier Mapping section is therefore a plurality of encoded carrier signals, each at a respective carrier frequency according to the transmission scheme. Data from one or several sources is typically split across different carriers. Pilot information needed to synchronise the receiver is also provided by Pilot Signal Generator 20 to the Carrier Mapping section 18. In the example shown, it is assumed that the Carrier Mapping section mixes the pilot information with source data on specific carriers -the second and fourth carrier signal in the diagram -rather than transmitting the pilot information only on specially reserved pilot carrier signals. In practice however, and with the provision of appropriate switches, the pilot data could be carried on any or all of the carriers.

In the frequency domain, the phase delay discussed above is introduced by multipliers 22. For each carrier signal of frequency f, a multiplier 22 modifies the carrier at a frequency f by multiplying the signal by the appropriate complex number needed to achieve the desired delay for that carrier. The multiplier applies the multiplication factor given by the expressions for H(f) above. Across the entire frequency spectrum of the carrier signals, the effect of the respective delays will be to produce the spread shown in Figures 4 to 6.

The second and fourth signal paths, whfch are arranged to occasionally carry the pilot information, are shown as including switches 24 controlled by the Pilot Generator 20. Normally, the switches are in the position shown in the diagram, that is the switches are positioned so that the delay occurs for those carriers. In combination with the delays present on the other carrier paths, this means the each of the paths are delayed according to their frequency. However, it is preferable to periodically send un-rotated pilot information, and when this is occurring, pilot information is passed to the Carrier Mapper for inclusion in either of the second or fourth carrier, and the pilot generator operates the switch for that carrier to exclude the delay adder. This bypasses the phase rotation for the carriers In practice, the pilots could be added to any of the carrier signals and there are switches in each path accordingly. It will be appreciated that these carriers will be chosen to correspond to the special, un-rotated pilots on that particular symbol.

It is preferable to send both rotated and un-rotated pilot information. The time synchronisation algorithm in the receiver is likely to be confused by rotated pilots as the apparent delay will be different to the true delay, and so requires un-rotated pilots. The potential confusion of the time synchronisation was discussed above in reference to Figure 3. The channel equalisation algorithm however will need to see the rotated pilots, as these better represent what has happened to the data cells during transmission. As a result of this the example transmitter preferably broadcasts two sets of pilot information on the carrier signals: one that is rotated, and one that is not. The non-rotated pilots could be sent less often that the rotated ones since they are only needed for time synchronisation, reducing the extra overhead If only one set of pilot information can be transmitted, because of restraints on data capacity, then rotated pilots are preferred because of the severe consequences that mis-equalised data has at the receiver. It will be appreciated that the rotated and non-rotated pilots could be transmitted anywhere within the transmitted signal providing there is agreement (at the design stage) between the transmitter and receiver.

The parallel carrier signals are then received by the Inverse Fourier Transform block 26, and converted into a single multi-frequency signal for transmission.

Finally, the guard interval adder block 28 adds the guard interval to the signal.

The signal is then passed to a front end for transmission.

In practice, the proposed frequency response has two parameters: tmax and f0, representing the maximum group delay and the pitch of repetition in frequency respectively. The optimum choice of these parameters is an area for further study. To generate an example response, values were chosen as follows. The maximum group delay trnax was set to lOus by analogy with the value chosen for cyclic delay in simulations already conducted.

Bearing in mind that the group delay is a sine function, we should choose f0 so that there is less than half a cycle across the occupied bandwidth: this will avoid the group-delay becoming flat anywhere in the useful bandwidth, which would reduce the likelihood of cancellation with another signal.

It is possible to choose the values such that the zero-delay term H0 disappears: this will happen if Tmax. 10 x0, x0-2 405. In our example, this gave a value for f0 of 240.5kHz.

This was thought to be a good idea because it would allow a simple signal to be transmitted from the other antenna with zero delay without it interfering destructively with one of the equivalent paths from the first transmitter, although this is unlikely to be done in practice for the reasons discussed above, and there might be better ways of choosing f The magnitudes of the Bessel terms, i.e. the strength of each path in the impulse response, are given in the following table, along with the one-sided and total delay.

n J,, Delay (us) linear dB total one-sided 0 0 -ci) 0 0 1 0.519 -5.69 8.3 4.2 2 0.432 -7.29 16.6 8.3 3 0.199 -14.02 249 12.5 4 0.0648 -23.77 33.3 16.6 0.0164 -3571 41.6 20.8 6 0.00341 -49.35 49.9 24.9 The terms do indeed decrease rapidly in magnitude. A commonly used approximation (that leads to Carson's rule for Frequency modulation) is that J(x) is small provided n > x+1; in our case x0=2.405 so we would ignore the terms after n=3.

Figure 9 shows the group delay as a function of frequency for the parameters discussed. Curves are shown for a range of values of N, the number of pairs of Bessel terms included Including more terms increases the delay spread: each pair of terms extends the delay spread by 8.3us, but in return the group delay profile becomes a better approximation to the intended sine wave. Although the Carson's rule" approximation suggests we could use N=3, the figure suggests that N=4 would be a better choice.

Figure 10 shows the corresponding frequency response magnitude as N varies.

The passband ripple is fairly inconsequential (less than 1dB) for values of N of 3 and above, and insignificant for 5 and above.

Note that, because of the parameter choices, this is only about 40% of a cycle of the sinewave. Consequently it does not reach the maximum group delay of lOus, but consequently it remains relatively linear without flattening off seriously Any of the above profiles will give an impulse response that appears to be centred on the true delay value, such that the receiver's time synchronisation will behave sensibly. What kind of fading will occur if a receiver can see signals from two transmitters? We have decided that we want a similar shape of impulse response from the two transmitters However, if the same phase response is applied to both, the likelihood of flat fading will be exactly the same as if nothing had been done to the signals. A possible solution is to apply the same group delay profile to both, but with the opposite signs of group delay. In other words the frequency response applied to one transmitter would be the conjugate of that in the other case.

The way in which the signals from the transmitters combine will depend on three things: the relative delay, the phase and amplitude of the two signals. Since the frequency response we are applying is almost constant amplitude we can assume that the worst case for flat fading is when the signals have equal amplitudes.

The total power at the receiver was calculated for a range of relative delays and phases for the parameters given above and for N6. The results are plotted in Figure 11: the vertical axis shows the total received power in dB relative to the power of a single transmitter whilst the horizontal axis gives the relative delay.

The different curves explore the possible phase differences in 64 steps. The worst fading happens between zero delay and about lOus. However, the total received power is never significantly less than for one transmitter alone, so that switching on the second transmitter would never make reception significantly worse. Note however that the worst case represents a reduction of 3dB compared to the total radiated power.

As the delay gets longer the variation with relative phase becomes less extreme and the signals tend to combine by power addition, giving a +3dB combined signal power.

Figure 12 shows the same plot for the linear group-delay variation (i.e the chirp function) The worst case here is slightly worse, with the combined power being a little over 1dB less than for a single transmitter.

A family of frequency responses has therefore been found which can be applied to the OFDM carriers, with the following properties: 1) The corresponding impulse response is of bounded extent (delay spread); 2) The frequency response is nearly constant in amplitude; 3) The total impulse response extent can be traded against passband flatness, 4) The apparent impulse response will be symmetrical about the true delay, so that the time synchronisation in the receiver can respond correctly to it; 5) Two signals can be generated with opposite signs of phase and transmitted from two different transmitters in a network; and 6) Two signals generated in this way, but with opposite signs of group delay, cannot suffer significant destructive interference whatever the relative delays and phases; in the worst case combination, the received total signal power will be approximately equal to that of one of the two transmitters.

The best parameter values are likely to depend on the specific application of the invention, since there are two fundamental parameters to be adjusted in addition to N, the number of Bessel terms to be included Figure 13 is a corresponding transmitter block diagram for apparatus that adds the delay in the time domain, rather than the frequency domain. The structure of the apparatus is the same as Figure 8, only the adders 22 and the switches 24 are omitted, and replaced with corresponding cyclic delay blocks 30 and gain multipliers 32. These feature after the lEFT block 26 in the time domain.

The time domain signal is therefore passed to each of the delay blocks, and delayed by a different amount. The differently delayed copies of the signal are then multiplied by respective gain multipliers 32 and recombined by adder 34, thereby creating the delay-spread signal shown in Figures 4 to 6. The gain applied to the delayed versions of the signal, ensues that the zero delay version of the signal has the maximum amplitude. In order to do this, it may be necessary to move some copies of the signal forward in time by applying a negative delay, or by delaying all of the copies and applying maximum gain to the copy in the middle of the spread, so that it appears at the receiver to be the undelayed version The signal is then passed to Add Guard Interval block 28 as before. As noted earlier, where antenna diversity is used, both antennae preferably spread the signal in this way, however the spread may also be applied at just one antenna or the other Figure 14 shows a block diagram for an example OFDM demodulator according to the invention. First, an analogue signal is received by the antenna 36 and fed to the RF Front End 38, which takes the frequency of the signal down to a base band frequency for processing The base band signal is then fed to Analogue to Digital Converter 40 to produce a bit stream of the received signal, which is passed to channel filter 42. The output of the channel filter 42 is fed to the time synchronisation block 44, which also receives an input from the later TFFT block 58 The signal is then passed to FFT 46 and to Automatic Freq Control block 48.

The frequency domain values after automatic frequency control block 48 are partitioned by block 50 into data and pilots as normal, but the pilots are further partitioned into those that were phase rotated at the transmitter and those that were not The receiver is able to do this as the location of the data, pilots, and non-rotated pilots will have been fixed by the transmission standard. The rotated pilots are fed to the Frequency and Time Interpolation block 52 which generates an estimate of the channel frequency response on every symbol and carrier. This estimate is used to equalise the received data constellations which also underwent rotation at the transmitter at the equaliser 54 The non-rotated pilots are also used to generate an estimate of the channel frequency response by a second Frequency and Time Interpolation block 56, but in this case there will be no rotation and the true response will be measured. This is passed to an Inverse FFT block 58, whose output is an estimation of the impulse response. This estimate is fed to the time synchronisation process 44, which uses it to optimise the timing according to the various known algorithms.

The data is also passed to the equaliser 54. The output of the equaliser is passed to the Channel Decoding and Interleaving block 60.

Although the technique has been discussed in the context of closely-spaced transmit diversity for DRM , it is also applicable to other COFDM-based systems including the existing DRM standard for use below 30MHz. It could be particularly beneficial in more conventional SFNs with a greater transmitter spacing, because of the way in which it avoids flat fading even when the signals are co-timed.

The system could be implemented in software, in hardware or in a combination of both

Claims

Claims 1. A transmitter comprising: a first and second antenna arranged

to transmit a multi-carrier signal in an antenna diversity scheme: cyclic delay means connected to at least one of the first or second antennas and arranged to introduce a spread of cyclic delays into the signal such that portions of the signal are delayed by different amounts before transmission.

2. The transmitter of claim 1 wherein the delay means is arranged to introduce a frequency dependent phase delay into the signal.

3. The transmitter of claim 2 wherein the phase delay varies in a non-linear manner with frequency.

4. The transmitter of claim 3 wherein the phase delay varies sinusoidally with frequency.

5. The transmitter of claim 3, wherein the phase delay varies quadratically with frequency 6 The transmitter of any of claims 1 to 5, comprising a frequency-to-time domain converter for receiving input signals from a plurality of separate carrier signal paths, and combining the signals into a single multicarrier signal, each carrier signal having a respective frequency; and a multiplier in one or more of the signal paths for introducing the frequency dependent phase delay into the signal, according to the frequency of the carrier signal.

7 The transmitter of claim 6, wherein the maximum phase delay is applied to the carriers at the ends of the frequency spectrum.

The transmitter of claim 6 comprising: a pilot signal generator for adding a pilot signal into respective ones of the arate carrier signals; a switch associated with each multiplier in a signal path for by-passing the multipher, such that a delayed or a non-delayed pilot signal can be added to the carrier signal.

9. The transmitter of claim 6, wherein the multiplier modifies the carrier signal by superimposing a signal given by H(O H(f) = e11 where 0(J) = -2ir--and B is the bandwidth.

10. The transmitter of claim 6, wherein the multiplier modifies the carrier signal by superimposing a signal given by H(f).

H([) = where 0(f) = taJ cos(1] and tmax and f0 representing the maximum group delay and the pitch of repetition in frequency between the carriers.

11. The transmitter of claim 6, wherein the multiplier modifies the carrier signal by superimposing a signal given by H'(f) where

N

J-I'(f) = ,,N1"uh1m f0)e where Jn(X) is the Bessel function of the first kind, order n, and where Tmax and f0 represent the maximum group delay and the pitch of repetition in frequency between the carriers, and N determines how many pairs of Bessel functions are included.

12 The transmitter of any of claims 9, 10 and 11, wherein the carrier signal is divided into symbols in the time domain, and the multiplier superimposes the signal H(f) onto respective symbols 13. The transmitter of claim 1 comprising a frequency-to-time domain converter for receiving input signals from a plurality of separate carrier signal paths, and combining the signals into a single multicarrier signal, each carrier signal having a respective frequency; and a plurality of cyclic delay blocks arranged to each receive a copy of the signal from the frequency-to-time domain converter, and respectively delay the copy of the signal by different amounts, and an adder arranged to receive the delayed versions of the signal and combine them so that the signal has an apparent spread in the time domain.

14 The transmitter of claim 13, comprising a gain multiplier associated with each of the delay blocks, such that the envelope of the apparent spread in the time domain can be shaped 15. The transmitter of claim 14, wherein the gain applied to the delayed versions of the signal, ensues that the zero delay version of the signal has the maximum amplitude.

16 The transmitter of any preceding claim, wherein the delay means introduces delays in the range of 0 to 50 ps compared to the timing of a non-delayed version of the signal.

1 7 The transmitter of any preceding claim, wherein the delay means is arranged to introduce delays into the signal transmitted by both antennas.

18 The transmitter of claim 17, wherein the delay means applies different delay parameters at different antenna.

19 The transmitter of any preceding claim in which the transmission system is one or more of OFOM, COFDM, DRM, Digital Terrestrial TV, and/or DVB 20. The transmitter of any preceding claim wherein the frequency-to-time domain converter applies a Fourier Transform, an Inverse Fourier Transform, an FF1 or an IFFT.

21 A receiver for receiving a multicarrier signal from the transmitter of claim 8, comprising, a time-to-frequency domain converter for converting the time domain multicarrier signal into a plurality of separate carrier signals having respective carrier frequencies; a partitioning block for extracting delayed pilot signals, non-delayed pilot signals and data from the separate carrier signals; a first interpolator for receiving the non-delayed pilot signals, and generating a channel frequency response for time synchronisation; a second interpolator for receiving the delayed pilot signals and generating a channel frequency response for equalising the data, a channel equaliser for receiving the delayed data and the output of the second interpolator 22. A method of transmitting a multi-carrier signal in an antenna diversity scheme comprising introducing a spread of cyclic delays into the signal passed to at least one of the first or second antennas such that portions of the signal are delayed by different amounts before transmission.

23. The method of claim 22 wherein the cyclic delay is a frequency dependent phase delay.

24 The method of claim 23 wherein the phase delay varies in a non-linear manner with frequency The method of claim 24 wherein the phase delay varies sinusoidally with frequency 26 The method of claim 24, wherein the phase delay varies quadratically with frequency.

27. The method of any of claims 22 to 26, comprising receiving input signals from a plurality of separate frequency domain carrier signal paths, each carrier signal having a respective frequency; introducing the frequency dependent phase delay into the signal, according to the frequency of the carrier signal; and converting from the frequency domain to the time domain, and combining the signals into a single multicarrier signal 28 The method of claim 27, wherein the maximum phase delay is applied to the carriers at the ends of the frequency spectrum.

29. The method of claim 27 comprising: generating a pilot signal for addition into respective ones of the separate carrier signals; delaying or not delaying the pilot signal so that both a delayed or a non-delayed pilot signal can be added to the carrier signal The method of claim 27, wherein the step of introducing the frequency dependent phase delay comprises modifying the carrier signal by superimposing a signal given by H(f): H(f) = where (f) = and B is the bandwidth.

31. The method of claim 27, wherein the step of introducing the frequency dependent phase delay comprises modifying the carrier signal by superimposing a signal given by H(f): H(f) = where (2 0(f) TmaxfO COS Jo and tmax and f0 representing the maximum group delay and the pitch of repetition in frequency between the carriers 32. The method of claim 27, wherein the step of introducing the frequency dependent phase delay comprises modifying the carrier signal by superimposing a signal given by H'(f) where

N

H' (f) = I'n (Tma fo)e it, where Jn(X) is the Bessel function of the first kind, order n, and where tma and f represent the maximum group delay and the pitch of repetition in frequency between the carriers, and N determines how many pairs of Bessel functions are included 33. The method of any of claims 30, 32 and 32, wherein the carrier signal is divided into symbols in the time domain, and the superimposition of the signal H(f) is onto respective symbols.

34. The method of claim 22, comprising: receiving input signals from a plurality of separate frequency domain carrier signal paths, each,carrier signal having a respective frequency; converting from the frequency domain to the time domain, and combining the srgnals into a single multicarrier signal; and providing a plurality of cyclic delays to copies of the signal received from the frequency-to-time domain converter, to respectively delay the copies of the signal by different amounts; and receive the delayed versions of the signal and combine them so that the signal has an apparent spread in the time domain.

35 The method of claim 34, comprising.

applying a gain multiplier to each of the delays, such that the envelope of the apparent spread in the time domain can be shaped.

36 The method of claim 35, wherein the gain applied to the delayed versions of the signal, ensues that the zero delay version of the signal has the maximum amplitude.

37 The method of any of claims 22 to 36, wherein the delays introduced are in the range of 0 to 50 ps compared to the timing of a non-delayed version of the signal 38 The method of any of claims 22 to 37, wherein the delays are introduced into the signal transmitted by both antennas.

39 The method of claim 38, wherein the delay means applies different delay parameters at different antenna 40 The method of any preceding claim in which the transmission system is one or more of OFDM, COFDM, DRM, Digital Terrestrial TV, and/or DVB 41 The method of any of claims 22 to 40, wherein the frequency-to-time domain converter applies a Fourier Transform, an Inverse Fourier Transform, an FFToranIFFT 42 A receiving method for receiving a multicarrier signal from the transmission method of claim 27, comprising; converting the time domain multicarrier signal into a plurality of separate carrier signals having respective carrier frequencies; extracting delayed pilot signals, non-delayed pilot signals and data from the separate carrier signals, receiving at a first interpolator, the non-delayed pilot signals, and generating a channel frequency response for time synchronisation, receiving at a second interpolator, the delayed pilot signals and generating a channel frequency response for equalising the data; receiving the delayed data and the output of the second interpolator and performing a channel equalisation based on the data and the output.

43. A transmitter substantially as described herein with reference to the drawings 44. A receiver substantially as described herein with reference to the drawings; A transmission method substantially as described herein with reference to the drawings.

46. A reception method substantially as described herein with reference to the drawings