PRIOR ART It is well known from the literature that the performance of broadband communication systems with respect to necessary Signal to Noise Ratios (SNR) for a certain Bit Error Rate (BER) can be greatly improved by applying a narrow stop filter before the demodulation of the broadband signal takes place, if the input signal is corrupted mainly by a narrowband interference. For an introduction to this field, see for example the book by John D. Proakis: "Digital Communications"ISBN 0-07-05-0937-9, second edition, Mc Graw Hill 1989, chapter 8.3.5.

The general block diagram of such a system is given in fig 1. First the combined Signal (broadband user signal) and the Interference (unwanted narrow band disturbance) is entering a filter. This filter may be a tapped delay line filter. The coefficients of this filter are computed in such a way that the Interference is attenuated more than the Signal, i. e. the filter must have the form of a narrowband stop function around the frequency of the Interference. Most often the resulting SNR out of this filter is not appropriate for the demodulation of the information content of the Signal, and therefore a correlator, matched to the broadband Signal follows the filter. At the output of this correlator, the SNR has improved further, and the demodulation of the user information may take place.

PROBLEMS WITH PRIOR ART The computation of the parameters in the interference suppression filter is normally a process that requires a

large number of operations per sample of the signal. For this reason the methods described in the literature are not very suitable for implementation in circuitry for handheld, lightweight communication equipment. The large number of operations necessary means that either a large complicated integrated circuit or DSP has to be used, or the circuitry will be very power consuming. Also the large number of operations necessary will lead to a long adaptation time, not compatible with frequency hopping systems where the time available for adaptation is extremely short.

In most systems the parameters of the filter may only be calculated while the interference is present alone. If the user signal is present, it is impossible to calculate the parameters of the filter in such a way that the filter will improve the Signal to Noise ratio at the output to a level useful for demodulation of the signal.

OBJECT OF THE INVENTION An object of the invention is to provide a method for supplying a quick and accurate estimate of the filter parameters of an interference cancelling filter, and in particular to provide a filter and a method for calculation of the filter parameters that requires a minimum of operations per sample of the signal, making the system suitable for implementation in small, lightweight, battery operated equipment.

Another object is to make it possible to calculate the filter parameters while the user signal and the interference is present together, by applying a correlator after the filter that is matched to the user signal and therefore increases the SNR further, to a value useful for demodulation of the signal.

These objects are achieved in a filter and method according to the appended patent claims.

DESCRIPTION OF THE INVENTION The invention will now be described in reference to the appended drawings, in which: Fig. 1 shows a method for suppression of an interfering signal in a broadband communication system. A block diagram comprising a narrow band filter and a correlator is shown.

The spectrum of total signal at input and out of the narrow band filter is also shown. Finally the correlation peak in a noise background is also shown.

Fig. 2 shows a notch filter for suppression of an interfering signal according to the present invention.

The invention relates to using a transversal filter with only one variable coefficient as shown in figure 2. The filter is a simple three stage transversal filter. d denotes the sample delay i. e. the time between samples of the signal. Three subsequent samples are weighted and summed to provide an output signal y (t), where t is time.

This may be expressed as y (t)-x (t) + A*x (t-1) + x (t-2), which means that only one of the filter coefficients, A, is made variable, while the remaining coefficients are set to one. The filter is a special variant of a transversal filter, since the coefficient in the last path is always 1.

The filter is a notch filter, which always will give a perfect cancellation of one single sinusoid of a certain frequency. It is easily seen that if the coefficient A is made equal to a =- [x (t) + x (t-2)]/x (t-1), the output of the filter is always zero. This happens for the frequency given by the equation 2cos (2z*f*d) =-a.

This filter requires very little HW and/or very little calculation steps. The calculation of a requires only one division (may be realised by a table look-up) per sample.

However, if the coefficient a is used, the output is always zero. In order to let some of the user signal through the filter, we therefore perform a function on the a's : A = f (a). This function may be a simple averaging of a's before

and after the time t-1. The function must be such that the notch track the narrowband interference, but is too slow to track the broadband user signal. When averaging a's over a time interval T, it is essential that the coefficient A acts on the data sample in the middle of the averaging interval used for calculating A. This is achieved by introducing the delay m/2 ahead of the filter.

If the signal sample x (t-1) in any case should happen to be zero, the calculation of a will fail. In such a case, the system may ignore this particular a, and depend on the other a's in the calculation of A=f (a). A better method is to replace the particular a by an a calculated from three samples shifted one time interval d. This method is useful when the sample rate of the signal is large compared to the interference variation rate.

ADVANTAGES OF THE INVENTION A system using the filter in fig 2, and with the function f equal to an averaging of 8 a's, has been simulated on a computer. This system requires only 3 additions and one multiplication per sample to perform the filter function.

To calculate the A one addition and one subtraction is required per sample. Only 10 samples of the input signal are required to achieve the first useful sample of the filtered output.

Such a filter has been simulated with input consisting of Frequency Modulated Interference along with a Direct Sequence Spread Spectrum user signal. The results show that the communication link may be operated as if the Interference was not present, i. e. with satisfactory BER, in the following situation: FM deviation and modulation tone within practical limits for actual FM systems in use, I/S of up to 40 dB and with Analogue to Digital converters of practical number of bits. This represents a dramatic improvement compared to a system without a filter, and the filter system is dramatically simpler than any found in the literature.