GB2212033A - Time synchronisation - Google Patents

Abstract

A method and apparatus (Fig. 1) for synchronising the sampling frequency at reception of digital data utilises a timing correction algorithm which takes into account level changes in the signal due to fading or multipath propagation, by presuming that the relative change in level, due to fading, is essentially the same for corresponding samples, i.e:- Yo,i-1,h/Yo,i,h/ SIMILAR Yo,i-1,h-1/Yo,i,h-1 Use of this presumption in the otherwise known error algorithm ensures that a pure fade does not per-se introduce sampling frequency errors. The precise sampling points used in the algorithm may depend upon transmission conditions. <IMAGE>

Description

Time-Synchronisation This invention relates to time-synchronisation of digital data and in particular it relates to a method and apparatus for synchronising the reception of digital data transmitted over for example an HF radio link.

In the serial transmission of digital data over a radio link, serious difficulties are often experienced in achieving correct time synchronisation at the receiver.

It is clearly important that a sampling receiver keeps fully in synchronisation with incoming data in order that the data is correctly sampled. An HF radio link tends to introduce a combination of multi-path propagation and fading which leads to apparent variations in the rate of arrival of the digital signal at the receiver.

Various methods for synchronising a receiver have been developed and one such system is described by Clark and McVerry: "time synchronisation of an HF radio modem", IEE Proc. F, Communication, Radar & Signal Processing, 1982, 129, (6), page 403-410. in which a technique for achieving time synchronisation of a receiver for a quadrature amplitude modulated (QAM) signals operating at 9,600 bits per second is discussed. On an impulseresponse system, if there is a difference in timing between samples then the value of one point measured during two relative sampling times will be different and hence will lead to inaccuracies. The Clark and McVerry system adjusts the timing by measuring the slope between samples, which is taken to be the difference between the two samples and adjusting the timing accordingly.

However, one problem not addressed up to now has been the problem of a level change between samples, regardless of whether there is any timing change. A level change typically occurs when there is fading of a signal. As will be shown below, a level change between adjacent samples will be seen, by a prior art system, as a timing change and hence could cause timing to be wrongly corrected and hence to be made worse.

According to the present invention in a first aspect there is provided a method of sampling digital data by repeatedly sampling at a plurality of particular points on a recurring waveform wherein an error signal for altering the sampling frequency is adapted by assuming that a change in level of a first selected sample point between any two samples is proportional to the change in a second selected sample point between those two samples.

Preferably, the level change in the Hth component is calculated by assuming that it is proportionally the same as the change in the (H-l)th component on each sample (or waveform). The change may be measured between adjacent waveforms or samples, either the preceding or the directly following sample or could be the second, third, etc.

proceding or following samples. By using samples some time apart then greater level changes will be observed which can help if measurement apparatus of only limited arithmetic accuracey is being employed, which may not be able to adequately distinguish small level changes.

The magnitude of a level change taken into account by the system may be limited so as to ignore obvious measurement errors.

Furthermore, only certain parts of the sampled inpulse response need, in certain embodiments, be considered to take into account level changes and other parts can be ignored, as will be shown in further detail below.

The accuracy of obtaining level changes can be improved by using double sampling techniques, as are known in the art.

Embodiments of the invention will now be described by way of example only with reference to the accompanying drawings, in which; Figure 1 shows in schematic form apparatus for the transmission and reception of digital information, and Figures 2 and 3 are illustrative waveform diagrams.

Figure 1 shows schematically apparatus for transmitting and receiving data over an HF link by using a sampled-impulse response method with QAM signals. The apparatus is that used by Clark and McVerry in the above mentioned publication. The information to be transmitted is carried by the data symbols CS.I which are complex digital signals having real and imaginary parts such that S = S . + jS1 i = S ,. + 0,1 and

li = +/- 1 or +/- 3 and 5l,i = +/- 1 or +/- 3. Each of the sixteen possible values of [Si] are equally likely.The data symbols are applied to a low pass filter 1 and linear modulator 2 which convert the data symbols [Si] into a serial stream of signal elements, each element comprising the sum of two quaternary double side band suppressed carrier and amplitude-modulated elements with the carriers in phase quadrature; the "in phase" and "quadrature" elements being determined respectively by the real and imaginary parts of the respective data symbol. The concept of QAM signals is well known. The resultant sixteen level QAM signal, whose bandwidth is approximately 2,400 Hz, is fed to an HF radio link 3 where its spectrum is shifted into the HF band, the system is then transmitted via one, two or more independently Rayleigh fading skywaves to a receiver where its spectrum is returned to the voice band. White gaussian noise 4 can also be added to the data signals at the output of the HF link.A linear demodulator 5 includes a bandpass filter for passing only signals within the frequency band of the data signal. The thus filtered signal is fed to two linear coherent demodulators, which together form part of demodulator 5 in the Figure, whose reference carriers are in phase quadrature and have the same frequency. This frequency is equal to the average instantaneous frequency of the received signal carrier, thus eliminating any constant frequency offset in the received QAM signal but not tracking the small variations in the signal carrier frequency introduced by the HF radio link. The demodulated signals at the outputs of the in phase and quadrature coherent demodulators are respectively taken to be real and imaginary valued so that the resultant demodulated baseband signal R(t) is complex.

It is seen that the baseband signal generator, linear modulator, HF radio link and linear demodulator together form a time varying linear baseband channel. The data symbols (Si) can be considered to appear in the form of a sequence of impulses (Si i (t - iT)).

The waveform R(t) is sampled once per data symbol, at time instant (# i) where ideally # i = iT. the complex valued sample of the demodulated baseband symbol R(t), at time t = e i, is accordingly

and the sequence of complex values given by the vector i = (Yi,o Yi,l...Yi,g) is taken to be the "sampled impulse response" at time t =# # of the resultant linear baseband channel in Figure 1. It is assumed that Yi ,h = for h < 0 and h > g so that the sampled impulse response has a finite extent.

Figure 2 shows two typical impulse responses Y. (t) and Yi 1(t). An important property of the linear baseband channel is that it varies only slowly with time, and hence, as can be seen Yi-l and Y. do not differ significantly.

The received samples (Ri) are applied to an adaptive filter 6 and detector 7 which determines the detected data symbols (Si) ) at time =0i. At this time, a channel estimator 8 estimates the sampled impulse response of the channel for the received sample R. n' the delay in estimation being n sampling intervals, and then uses this and the previous estimates to predict the sampled impulse response Yi for the latest received sample Ri. A delay 9 is used to feed the appropriate received signal to the estimator. The article by Clark and McVerry referred to herein above describes how the sampling rate can be synchronised if there is a time difference between adjacent samples and accordingly this technique will not be described in detail.Summarising the error signal generated by the technique is given by

Where Yo,i,h and Yl,i,h are the real and imaginary parts respectively of i,h' and similarly mutatis mutandis, with Ui,h; where Uo,i,h = Yo,i,h+l - Yo,i,h-l; Ul,i,h = Yl,i,h+l - Yl,i,h-l. It is assumed that Yo,i,h = Yl,i,h for i < 0 and i > g.

A study of Figure 2 and a conceptual superimposition of waveforms Yi and Yi-l will show that if there is a delay A between two sample impulse responses then sampling at time T, 2T etc will give a false reading and somehow an estimate must be obtained for A such that the sampling times can be altered. The Clark and McVerry system achieves this but does not take into account a level change in the signal, since it is assumed in Figure 2 that the levels of the signals are identical.

Figure 3 shows a system in which there is a level change in a signal between the impulse-responses Y. and Yi-1. Such a level change can be introduced by multipath fading and propagation although it is to be noted that the technique to be described is only relevent to the case where the fading of two adjacent samples is coherent, i.e.

that in which all parts of the sampled impulse response fade at a proportional rate. It is seen from the figure that the sampled impulse response Yh-1 is at a reduced level compared to that of Yh In the Figure only a portion of each impulse response is shown. Since the previous Clark and McVerry system operates by building up an error signal e i from terms each of which is the change in successive sample values multiplied by the slope #Y #T estimated from the difference between nearest neighbours, then a study of Figure 3 will show that a difference in sample value will be perceived between samples, for instance Y and Y which difference will be o,i-l,h' perceived to be a difference in timing.The system will therefore alter the timing, even though it is seen in Figure 3 that the timing is actually correct at this point. Accordingly, timing errors will be introduced by applying a prior art system. The present system on the other hand takes into account these level changes by subtracting an estimate of the level change in the hth component, assuming that it is proportionally the same as the change in the (h-l)th component.Thus (Yo,i,h - Yo,i-l,h) # (Yo,i,h-l - Yo,i-l,h-l) Yo,i,h Yo,i,h-l which is equivalent to Yo,i-l,h - Yo,i-l,h-l Yo,i,h Yo,i,h-l and by subtracting the right hand side of the above equation from the left hand side, the error introduced by the level change is at least partly compensated for, By introducing this in the equation for the timing error in the Clark and McVerry paper we get

Yo ,1 n 2 U o,i,h-l + [(Yl,i,h - Yl,i-l,h) - (Yl,i,h-1 - Yl,i-l,h-l) Y, h U1 #@l,i,h# Yl,i,h-l# i.e. the effects due to level changes are subtracted from the original error signal to reduce the effects of fading on the accuracy of synchronisation.

It will of course be appreciated that it need not be the level change in the (h-l)th component which is used to obtain an estimate of the hth component; any other component such as the (h+l)th, (h+2)th, (h-2)th could equally well be used, although of course the greater the time interval between the two sample points the greater must be the time interval over which coherent fading is maintained, for correct operation to be achieved.

Furthermore, the level of change between samples need not be measured between adjacent samples Y; and Yi-l Clearly the level difference of a particular component between samples Yi and Yi k' where k > 1 will in general be greater than that between Yi and Yi-l' for a continually fading signal. This is particularly useful where computing apparatus of only limited arithmetic accuracy is being used, where the small changes occurring between immediately adjacent samples may not be detectable. As an example, if k=8 then the level change will be around 8 times that when k=l, which is equivalent to a 3 bit change in the required accuracy of the computing system. Values of k up to 16 have been successfully tested but k could feasibly be even greater than this value.

The level change in the signal which is taken into account by the system can be limited so that obviously false readings are ignored.

By using double sampling techniques the accuracy of slope measurements when using the timing algorithm can be increased. Such techniques are well known and will not be further described herein.

Referring to Figure 2, it should be noted that not all of the samples Yi Ox i,l Y. g need be taken into 1,g account when synchronising the timing algorithm. Indeed, it may often be found useful not to use some samples such as, in this example, Yi,4 - Yi,6 which have a smaller magnitude than samples Y. - i,3 However, samples 1,0 should only be ignored 'adaptively' since as sampling progresses the waveform can alter as, for instance, sky waves appear or fading varies, in which case it may become necessary to take different sampling points into account in order that level changes between samples are adequately compensated.

Claims (11)

1. A method of sampling digital data by repeatedly sampling at a plurality of particular points on a recurring waveform wherein an error signal for altering the sampling frequency is adapted by assuming that a change in level of a first selected sample point between any two samples is proportional to the change in a second selected sample point between those two samples.

2. A method as claimed in claim 1 wherein the level change in each selected component (hth component) is calculated by assuming that it is proportional to the change in the (h-l)th component.

3. A method as claimed in claim 1 or claim 2 wherein the level changes are measured between adjacent samples.

4. A method as claimed in claim 1 or claim 2 wherein the level changes are measured between samples other than immediately adjacent ones.

5. A method as claimed in any of the preceding claims wherein level changes of only selected sample points on the waveform are used to adapt the sampling frequency error signal.

6. A method as claimed in any of the preceding claims wherein only level changes up to a preselected threshold level are used to adapt the sampling frequency error signal.

7. A method as claimed in any of the preceding claims wherein the data is QAM data.

8. A method as claimed in any of the preceding claims wherein the error signal to be adapted is that which is herein before defined as

9. Apparatus adapted to perform the method of any of claims 1 to 8.

10. Apparatus as claimed in claim 8 and substantially as hereinbefore described with reference to, and as illustrated by, the accompanying drawings.

11. A method for sampling digital data substantially as hereinbefore described with reference to, and as illustrated by, the accompanying drawings.