DESCRIPTION OF THE RELATED ART A filterbank can be used to analyse the frequency contents of a signal by comparing the signal against a set of filter coefficients. One such filterbank is illustrated in Figure 1 of the accompanying drawings, and will be described in detail below. Each set of filter coefficients represents a particular frequency content and results in a particular output when applied to the signal. The frequency contents of the signal can then be determined by studying the outputs of all the filters in the filterbank.

Some applications, such as estimation of Doppler spread for a Mobile Station (MS) in a mobile telecommunications system, require the exact frequency contents of the signal to be determined. For these applications many filters are required to improve the frequency resolution of the filterbank.

SUMMARY OF THE INVENTION The present invention presents an interpolation method for frequency analysis of a signal using a filterbank.

The outputs of the filters are analysed and those sets of filter coefficients with frequency properties closest to the input are determined. These filter coefficients and their outputs are then used to

determine the frequency contents of the input signal.

According to one aspect of the present invention, there is provided a method of analysing frequency contents of an input signal received by a filterbank which comprises a plurality of multipliers connected to receive respective samples of the input signal and operable to multiply the samples by respective coefficients to produce a plurality of multiplied signals, and adding means operable to add the multiplied signals to give a filterbank signal, the method comprising : determining a first set of coefficients, which set gives the filterbank output having the largest magnitude ; determining a second set of coefficients ; interpolating between the first and second sets of coefficients to give a third set of coefficients, which third set is indicative of the frequency contents of the input signal.

BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 illustrates a filterbank ; Figure 2 is a graph illustrating an input signal and coefficients of the filters in the filterbank of Figure 1 ; Figure 3 illustrates vector notation of the signals of Figure 2 ; and Figure 4 illustrates steps in a method embodying the present invention.

DESCRIPTION OF THE PREFERRED EMBODIMENT

A filterbank for frequency analysis of a signal can be interpreted as a set of matched filters in the time domain.

Figure 1 illustrates a realization of a filterbank which can be used to analyse the frequency contents of an input signal, x (n). The filterbank comprises M multipliers Mo... MM-1 which operate to multiply samples x (n)... x (n-M+1) of an input signal by respective coefficients co (k)... CM_1 (k). The first multiplier Mo receives an undelayed sample x (n) of the input signal, and the succeeding multipliers receive a successive delayed samples x (n-l)... x (n-M+1) of the input signals.

The filterbank also comprises a plurality of M-1 adders al... a,,-, which operate to add together the outputs from the M multipliers. The filters are preferably provided by finite impulse response (FIR) filters having respective coefficients. The coefficients co... CM_1 are provided as a set : These coefficients are chosen with respect to frequency contents of input : Since the frequency contents of the input signal are not known, different sets of filter coefficients are employed in order to determine those contents. The frequency contents of the input signal are then estimated by identifying the set of coefficients that

give the largest output of the filterbank. Using matched filters to estimate the frequency contents is an optimal method in maximum likelihood sense if the additive noise of the input signal is white and has Gaussian distribution. However, the complexity of the method is large if the frequency estimate is required to have high resolution.

In a method embodying the invention, this complexity is reduced by first specifying those two sets of filter coefficients which produce filterbank outputs"closest" to the input signal x. The accurate coefficient values are then found by interpolating between these two sets of filter coefficients. One of these two sets is specified as being the set that gives the largest magnitude output from the filterbank. However, the second set of coefficients does not necessarily give the second largest output magnitude. Considering the sets of coefficients and the input x as vectors, the set of coefficients closest to x can be found.

Figure 2 shows, by way of example, an input signal x which is closer to the set of filter coefficients c (k- 1) than the set of coefficients c (k+1), even though the signal x is located between the sets of filter coefficients c (k) and c (k+1).

Therefore, it is necessary to study the relationship between two expressions :- (4)

For a better understanding of expressions (3) and (4), if x projects exactly onto c (k), then both (3) and (4) equal one. However, if the c (k) set does not map exactly to x, such that x is closer to the c (k+l) set (see Figure 2), then the numerator of expression (4) gets larger and so expression (4) becomes larger than expression (3).

The second set of filter coefficients can then be determined by selecting the set c (k-1) or c (k+1) that produces the larger of the expression (3) and (4). When the sets of filter coefficients are specified, the input signal frequency contents can be estimated by interpolating between the sets of filter coefficients.

Considering x, c (k), and c (k1) as vectors, the interpolation is performed according the following expression : (5) where a and ß are angles of (x, C (k)) and (x, #(k # 1)), respectively. The vector representation is illustrated in Figure 3.

Function relates a and ß to the frequencies represented by the filters c (k) and c (k1), and fk1 and fk are determined by the chosen sets of filter coefficients.

A method embodying the invention is illustrated in a flow chart of Figure 4, and comprises the steps of : Select the set of filter coefficients producing the largest filterbank output (Step A). This is projection of x on c (k) and is proportional to cos (a) in Figure 3.

Compare expressions (3) and (4) (Step B), in order to determine whether the signal x is located closer to c (k-1) or c (k+1).

If signal x is closer to c (k-1), compute Angle = (c (k), C (k-1)) jiT. C (k-1) and Ratio = - (Step C)<BR> XT.C(k) #If signal x is closer to c (k+1), compute Angle = (c (k), C (k +1)) XT. c (k + 1) and Ratio = XT C(k). (Step D) Initialise a and by setting each of them to half of Angle, as calculated. (Step E) Successively subtract a predetermined value from a

and add the same value to ß until Ratio #<BR> Cos(ß) The predetermined value is a small value which is determined by the resolution required by the filterbank.

# Estimate frequency spread by using updated α and ß values in expression (3) (Step G).

It will be appreciated that the invention can be employed, for example, in Doppler spread estimation, particularly where the radio channel is assumed to have Jakes spectrum. The channel correlation is estimated and analysed by a set of filter coefficients to determine the Doppler spread frequency.