GB2410873A

GB2410873A - Adaptive and constrained weighting for multiple transmitter and receiver antennas

Info

Publication number: GB2410873A
Application number: GB0402794A
Authority: GB
Inventors: Keith Wilson
Original assignee: Nortel Networks Ltd
Current assignee: Nortel Networks Ltd
Priority date: 2004-02-06
Filing date: 2004-02-06
Publication date: 2005-08-10
Also published as: GB0402794D0

Abstract

In a systolic ("heartbeat-like") weighting array 500, matrices X of input data from multiple receiving antennas (101, fig. 1) are manipulated by matrices C of spatial constraints at a processor (103) so that a beamformer (105) can extract wanted signals from multiple transmitters (eg. several GPS satellites). Narrowband (number of taps B =1, fig.2) or broadband B > 1, fig. 11) systems can thus detect multiple wanted signals without requiring multiple processors or square-root operations. When operating in an adaptive mode (L=1), an unconstrained solution is output which reduces interference omnidirectionally. Then, in a non-adaptive 'frozen' mode (L =0), spatial constraint matrices are applied which allow greater weightings from particular look directions, effectively steering a beam towards the sources. Transposing and passing the outputs back through the array further optimises the calculated weightings.

Description

J

24 1 0873

SIGNAL PROCESSING METHOD

FIELD OF THE INVENTION

This invention relates to methods and apparatus for detection of wanted signals in an environment where interfering signals are also present The invention also relates to improved signal processing methods for detection of wanted signals.

The invention particularly relates to radio signals, such as GPS (Global Positioning System) signals

BACKGROUND TO THE INVENTION

l0 In order to improve the reception of wanted signals in a radio network, it is known to use adaptive antenna techniques. Adaptive antennas use advanced signal processing and antenna array technologies to enhance the reception of the wanted signal In the presence of interference and noise. The interference could be signals from other radio networks, deliberate interference (jamming signals) or non-deliberate interference from radiation not associated with communications (e 9. microwave ovens). Noise may result from many sources including, but not limited to, the non-perfect nature of the radio channel, thermal noise in transmitters and receivers and atmospheric conditions.

An example of receiving apparatus is shown in figure 1. A signal is received on each of four antennas 101 resulting in 4 streams of data Z. Z4. This data is in digital form, having been demodulated and sampled by a receiver 107. The four signals, and therefore data streams, will not be identical and each signal will contain the wanted signal plus noise and interference (if present). The antennas are connected to a beamformer 105, and between each antenna and the beamformer is a tap 102 which taps off a representation of the signal and directs it to a processor 103. The processor therefore receives the data streams Z. - Z4 and can process the data to determine the way in which the streams should be combined in the beamformer in order to optimally extract the wanted signal from the noise and interference. The processor outputs parameters, known as weights (which may have complex form) which are fed 104 to the beamformer, which the beamformer then uses to extract the wanted signal from the data streams that it has received.

The optimum solution for the beamformer weights in a situation where the wanted signal direction is available is well known, and involves the inversion of the data covariance matrix formed from the received signals. Direct computation methods for the calculation of a matrix inverse can exhibit numerical instability, and so several known alternative techniques based on matrix decomposition of the data signal matrix have been developed. One such method is called QR decomposition.

In QR decomposition, a block of signal samples from a particular period of time is represented as a matrix, with each time sample being a row, and each antenna being a column. This matrix is factorised into a product of two matrices, referred to as the Q and R matrices. The matrix Q is unitary, i.e. when multiplied by its own conjugated transpose, the result is the identity matrix. The matrix R is upper triangular, i.e. all elements below the leading diagonal are zero. The calculation of the optimum weight solution is thus reduced to the inversion of a triangular matrix, which is a relatively simple computational task. The process does not explicitly require the formation of the data covariance matrix, reducing the dynamic range requirements of the computation, and thereby improving numerical robustness.

If the direction of the wanted signal is known, the solution can be constrained such that the antenna array gain is protected in this direction, in effect steering a beam toward the wanted signal. Mathematically, this is achieved by forming the matrix S. which is defined to be S=(R')HC, and then calculating the weight vector as W=R-'S. C represents a matched filter to the wanted signal, i.e. the complex conjugate of the wanted signal spatial signature. The H operator represents the conjugated transpose (Hermitian transpose).

A further key known advantage of QR decomposition is that it can be implemented with a systolic array. This approach uses repeated blocks of identical circuitry, and so allows a simple, scaleable architecture to be employed.

Although the systolic array-based QR decomposition technique described above works in the situation where there is one wanted signal which requires detection, the technique is not optimum for the situation where there are multiple wanted signals, for example in a GPS receiving system, where signals are required from a number of satellites (typically 4-8 satellites). In order to detect multiple wanted signals it is necessary to have multiple processors, one performing the weight calculation for each wanted signal, which increases the cost and size of processing capability required.

OBJECT TO THE INVENTION

The invention seeks to provide an improved signal processing method, in particular for calculation of weights for use in a beamformer, which mitigates at least one of the problems of known methods.

SUMMARY OF THE INVENTION

According to a first aspect of the invention there is provided a method of calculating weights for use in extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, said method comprising the steps of: feeding said plurality of received signals into a processor array; manipulating said array according to a plurality of spatial constraints, producing a plurality of output matrices, each one of said plurality of output matrices corresponding to one of said plurality of spatial constraints; reorientating said array and each one of said plurality of output matrices, producing a reorientated array and a plurality of reorientated output matrices; feeding back each one of said plurality of reorentated output matrices through said reorientated array; and outputting a plurality of weights from said reorientated array, each one of said plurality of weights corresponding to one of said plurality of reorientated output matrices, wherein each of said plurality of weights corresponds to a wanted signal.

Advantageously, this removes the interferers whilst maximising the gain of the wanted signals.

Advantageously, this provides an efficient implementation for the situation where there are multiple wanted signals which must be detected simultaneously.

The technique minimises the processor requirements, saving space and cost.

This is further assisted by the fact that the weights are calculated without requiring square roots. Floating point square root calculators implemented in FPGAs are relatively large and slow.

Advantageously, the method calculates explicit weights which can be extracted and used in an external beamformer unit. - 4

The step of feeding said received signals may comprise: passing said plurality of received signals through a processor array in a first adaptive mode, said array comprising cells arranged to perform predetermined operations.

Each of said plurality of spatial constraints may comprise a matrix, and the step of manipulating may compose: passing sequentially each of said plurality of spatial constraints through the array in a second nonadaptive mode; and storing a plurality of output matrices, each of said plurality of output matrices corresponding to the passing of one of said plurality of spatial constraints.

The step of reorientating may comprise the steps of: reversing the order of the lo rows in each one of said plurality of output matrices, producing a plurality of reorientated output matnces; and transposing cells in said array, producing a reorientated array.

The step of feeding back may comprise the step of: passing sequentially each one of said plurality of reorientated output matrices through said reorientated array in a second non-adaptive mode.

Each one of said plurality of spatial constraints may be associated with a wanted signal.

Each one of said plurality of spatial constraints may be the Hermitian conjugate of a spatial constraint matrix associated with one of said wanted signals.

The method may further comprise the steps of: A. feeding a first of said plurality of output matrices into said processor array; B. manipulating said array according to a temporal constraint, producing a first output matrix; C. reorientating said array and said first output matrix, producing a second reorientated array and a first reorientated output matrix; D. feeding back said first reorientated output matrix through said second reorientated array, producing a first output vector of a plurality of output vectors; repeating steps A - D for each of said plurality of output matrices, producing said plurality of output vectors, each one of said plurality of output vectors corresponding to a wanted signal; and multiplying each of said plurality of weights by the one of said plurality of output vectors corresponding to a same particular wanted signal, to produce a plurality of broadband weights, wherein each of said plurality of broadband weights corresponds to a wanted signal.

Advantageously, this technique allows wideband look-direction constraints with an insignificant overhead in terms of processing speed and FPGA resources compared to the unconstrained case.

Step A may compose: resetting said processor array to produce an initialised array, and passing a first of said plurality of output matrices through said initialized processor array in a first adaptive mode.

Step B may comprise: passing said temporal constraint through the array in a lo second non-adaptive mode; and storing a first output matrix.

Step C may comprise: reversing the order of rows in said first output matrix to produce a first reorientated output matrix; and transposing cells in said array, producing a second reorientated array.

Step D may comprise: passing said first reorientated output matrix through said second reorientated array in a second non-adaptive mode; and producing a first output vector of a plurality of output vectors.

According to a second aspect of the invention there is provided a method of extracting a plurality of narrowband wanted signals from a plurality of received signals comprising the steps of: calculating weights as described above; passing said plurality of weights to a beamformer; and combining said plurality of received signals according to said weights in said beamformer to extract said plurality of narrowband wanted signals.

According to a third aspect of the invention there is provided a method of extracting a plurality of broadband wanted signals from a plurality of received signals comprising the steps of: calculating broadband weights as described above; passing said plurality of broadband weights to a beamformer; and combining said plurality of received signals according to said broadband weights in said beamformer to extract said plurality of broadband wanted signals.

According to a fourth aspect of the invention there is provided a method of receiving a plurality of wanted signals comprising the steps of: receiving a plurality of received signals using a plurality of antenna elements; and extracting said plurality of wanted signals as described above. - 6

According to a fifth aspect of the invention there is provided a processor for calculating weights for use in extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, said processor comprising: an input for receiving said plurality of received signals; a processor array; means for passing said plurality of received signals into said array; means for manipulating said array according to a plurality of spatial constraints; means for storing plurality of output matrices resulting from said manipulation, each one of said plurality of output matrices corresponding to one of said plurality of spatial constraints; JO means for reorientating said array and each one of said plurality of output matrices, producing a reorentated array and a plurality of reorientated output matrices; means for feeding back each one of said plurality of reorientated output matrices through said reorientated array; and an output for outputting a plurality of weights from said reorientated array, each one of said plurality of weights corresponding to one of said plurality of reorientated output matrices, wherein each of said plurality of weights corresponds to a wanted signal.

According to a sixth aspect of the invention there is provided a receiver apparatus for extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, comprising. a plurality of antenna elements, each said antenna element for receiving one of said plurality of received signals; a processor as described above; and a beamformer for combining said plurality of received signals according to said plurality of weights.

According to a seventh aspect of the invention there is provided a radio 2s communications network comprising a receiver apparatus as described above.

According to an eighth aspect of the invention there is provided an adaptive antenna apparatus for receiving a plurality of GPS signals comprising a receiver apparatus as described above.

According to a ninth aspect of the invention there is provided a method of extracting GPS signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of GPS signals and a jamming signal, said method comprising the steps of calculating weights as described above; passing said plurality of weights to a beamformer; and combining said plurality of received signals according to said weights in said beamformer to extract said plurality of GPS signals.

The received signals may comprise a combination of GPS signals and a plurality of jamming signals.

The method may be performed by software in machine readable form on a storage medium.

The preferred features may be combined as appropriate, as would be apparent to a skilled person, and may be combined with any of the aspects of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

An embodiment of the invention Will now be described with reference to the lo accompanying drawings in which: Figure 1 is a schematic diagram of receiving apparatus; Figure 2 is a schematic diagram of a narrowband receiving apparatus; Figure 3 is a schematic diagram of a communications network containing the receiving apparatus of figure 2; Figure 4 is a flow diagram of the operation of a processor within the receiving apparatus of figure 2; Figure 5 is a schematic diagram of an array processor; Figure 6 is a schematic diagram of a boundary cell of the array processor of figure 5; figure 7 is a schematic diagram of an internal cell of the array processor of figure 5; Figure 8 is a further schematic diagram of the array processor of figure 5; Figure 9 is a further schematic diagram of the array processor of figure 5; Figure 10 Is a further schematic diagram of the array processor of figure 5; Figure 11 is a schematic diagram of a broadband receiving apparatus; Figure 12 is a schematic diagram of a communications network containing the receiving apparatus of figure 11; 8 Figures 13-15 is a flow diagram of the operation of a processor within the receiving apparatus of figure 11; Figure 16 is a further schematic diagram of the array processor of figure 5; Figure 17 is a further schematic diagram of the array processor of figure 5; Figure 18 is a further schematic diagram of the array processor of figure 5.

Figure 19 is shows a representation of a typical adapted beam pattern; and Figure 20 shows a representation of an example unadapted combined effective beam pattern.

DETAILED DESCRIPTION OF INVENTION

In Embodiments of the present invention are described below by way of example only. These examples represent the best ways of putting the invention into practice that are currently known to the Applicant although they are not the only ways in which this could be achieved.

In a GPS system, GPS receivers need to detect signals from more than one satellite in order to determine its location. Typically, a GPS receiver needs to detect at least 4 signals and preferably many more. in some circumstances, one or more jamming signals will intentionally be broadcast by third parties in order to hinder the GPS receiver, in addition to signals being subject to interference from many sources and also be subject to noise.

The wanted signals may be narrowband or broadband. The term 'wanted signal' is used herein to refer to a signal, substantially free from noise and interference, which is transmitted by another source. This is to differentiate the signal that is required, from the 'received signal' which contains a combination of the wanted signal(s), noise and interference.

The narrowband case is discussed below with reference to figures 2-10 and the broadband case is discussed below with reference to figures 5-18.

Although for many applications, a GPS signal can be considered as narrowband, in severe jamming environments, high cancellation performance of >60dB may be required In order to obtain this degree of cancellation, a broadband adaptive processor is essential. This can take the form of the tap-delay beamformer discussed below, or alternatively digital filtering can be applied to sub-divide the - 9 - passband into narrow sub-bands and the narrowband solution can be applied to each sub-band in turn.

An example of a narrowband receiving apparatus 200 is shown in figure 2, and its operation is discussed, by way of example only, for the situation where there are only two wanted signals (in reality there are likely to be more than two wanted signals). These wanted signals are two different signals 301, 302 originating from two different satellites 303, 304 as shown in figure 3. The system is subject to jamming interference signals 305. A signal is received on each of four antennas 201 resulting in 4 streams of data X' - X4. This data is in lo digital form, having been demodulated and sampled by a receiver 207. Each signal received includes a combination of wanted signals, noise and interference.

The four signals, and therefore data streams, will not be identical, as each wanted or interfering transmission arriving at the array will have a different spatial signature (the relative phasings of the signals across the antenna array being a function of the direction of arrival). The antennas are connected to a beamformer 202, and between each antenna and the beamformer is a tap 203 which taps off a representation of the signal and directs it to a processor 204. The processor therefore receives the data streams X' - X4 and processes the data to determine the way in which the streams should be combined in the beamformer in order to extract the two wanted signals from the noise and interference. In order to process the data, the processor may require information relating to the location of the satellites (described in more detail below) and this information may be obtained from a memory or database 205, or through knowledge of the position of the satellites relative to the antenna mounting, or through the use of known techniques for determining the direction of arrival, or spatial signature, of a transmitting source. The database may be co-located with the processor or may be located elsewhere and connected by a communications link. The processor outputs two vector parameters W., W2, one for each wanted signal 301, 302, known as weights (which may have complex form) which are fed 206 to the beamformer. The beamformer then uses W. to combine the data streams X, X4 to extract the first wanted signal and W2 to combine the data streams X, - X4 to extract the second wanted signal. The description below describes the processing activity within processor 204 In more detail, with reference to the process steps shown in the flow diagram of figure 4.

The adaptive algorithm within the processor uses a systolic array implementation. A systolic array, by analogy with the regular pumping of blood - 10 by the heart, Is defined as an arrangement of processing cells where data flows synchronously across the array between neighbours, with each cell performing a similar set of operations on the input data.

The algorithm utilises a triangular systolic array 500, such as that shown in Figure 5. The array has as many columns as there are beamformer elements NB, where NB is equal to the number of antenna elements multiplied by the number of delay taps per element. In the example shown, we have four antenna elements and, as this is a narrowband beamformer, only 1 tap per element.

Thus NB = 4 and the array has 4 columns as shown in figure 5.

lo There are two main types of cell within the array, referred to as boundary and internal cells, with each performing a different set of operations (described below). In figure 5, the larger circles 502 represent boundary cells, whilst the internal cells 504 are square in shape. The smaller circles 506 represent delay blocks - no operations are performed in these, but the data is delayed by one cycle period. The number of each type of cell required is simply given by Nboundary = NB Ninternal = (NB2 NB) / 2 Each cell contains a single stored complex value. Following convention, these have been labelled it, for the boundary cells, and Rid for the internal cells. Initially, the array is reset (step 401), with each of the stored values in the cells being set to zero.

The inputs and outputs of the boundary cells 502 are shown in Figure 6, for the kth array cycle. The cell has three inputs. For the first cell, these inputs are the sample data for the first element (x,n), the array 'freezing' control L'n (en), and a scaling factor An. The scaling factor is generally set equal to 1 unless otherwise described. For subsequent boundary cells, the inputs are connected to the outputs of preceding cells.

The operations performed by the boundary cells are detailed below: - 1 1 (k-l) Win (k) d =+Xin (k) Xin(k) {in(k) 4) | t Bin(k) E | d - (out(k) = t 0 (out(k) = 0 S(k) = t Din (k) x0ut(k) = Xin(k) Rout (k) = Akin (k) im(k) = {Xin(k)-d The value [m is stored within the cell, and updated each cycle. As shown above, the set of operations in adaptive mode (L'n = 1) is different from the set of operation in frozen, or non- adaptive mode (L'n = 0).

The inputs and outputs of the internal cells 504 are shown in Figure 7. The xn and S'n inputs are fed from the boundary cell, whilst the Yin input is connected to the input data samples, for the first row of cells; or the preceding cells, for subsequent rows.

The operations performed by the internal cells are detailed below: Rout (k) = Xin (k) SOut (k) = Sin (k) Yout (k) = Yin (k)-R(k - 1) Xin (k adaptive mode it(k) = R(k - 1) + Sin (k) Yout (k) frozen mode lo it(k) = R(k -1) As shown above, the set of operations in adaptive mode (L'n = 1) is different from the set of operation in frozen, or non-adaptive mode (L'n = 0) - 12 lnput data is then applied at the top of the array 500, (step 402), with the samples staggered in time by one cycle per element, as shown in figure 5. This staggering of samples is required to ensure that each sample wavefront, indicated by the shaded lines 508 in the figure, exercises the array of cells in the correct order.

The number of data samples in the block that is input to the array is determined by a number of factors, for example weight jitter considerations. Typically, several hundred samples will be used for each weight solution.

By passing the block of data samples through the systolic array, with the array operating in adaptive mode, the processor has formed the unconstrained lo solution. Output power has been minimised, and thus interference reduced, but no account has been taken of the wanted signals. Forming a constrained solution, in which the processor attempts to simultaneously cancel the interfering signals whilst protecting the antenna array gain in the desired look direction, requires further computational effort. By protecting the antenna array gain, the processor is effectively steering a beam towards the satellite where the wanted signal is coming from.

One of the advantages of the invention is that the look direction constraints, also referred to as spatial constraint matrices, (which implement the aforementioned beam steering) can be applied after adaptation to the input data; an important approach when separate weight solutions are needed for receiving multiple wanted signals from multiple sources. This minimises both processor resources and the number of processing operations required. As there are a very large number of data samples used (as described above, this could be several hundreds of samples), if the spatial constraint matrices are applied after adaptation to the input data, the input data only needs to be processed once, which takes for example, 1000 cycles for 1000 data samples. Application of the constraint matrices only takes a relatively small number of cycles in comparison (the number of cycles per matrix being equal to the number of delay taps per element, which is 1 for the narrowband case).

After the systolic array has adapted to the input data, the stored values in the cells contain a scaled version of the upper-triangular matrix R (as shown in figure 8). The boundary cells contain the square of r,,, whilst the internal cells have been scaled by r,' In figure 8 the beamformer is shown to have just four elements for illustration purposes only. The fact that the array does not simply contain R is a result of the fact that a square- root free algorithm has been employed. However, - 1 3 the benefits of not having to perform square roots outweigh the minor inconvenience of dealing with the scaling factors.

To ensure that the array does not continue to adapt, when it is not required to in future operations through the processor, the input L'n is used to freeze the stored values in the cells. When data samples are being inputted, L'n is set to '1', and the systolic array adapts (i.e. it operates in a first adaptive mode). When the look direction constraint matrices are being inputted, the array is frozen by setting Ln to 'O' (i. e. it operates in a second non-adaptive mode).

In the narrowband case, the spatial constraint matrix and its corresponding to output have only one column, and so are effectively vectors. In the first pass through the systolic array (step 403), following the adaptation of the array to the data, the complex conjugate of the spatial constraint vector associated with first wanted signal, C', is input. The resulting output (step 404) is a scaled version of the vector S., where Sit is equal to the product of the Hermitian inverse of R and C', as described in the Background section above. We label this output So. The process is shown in Figure 9, where s', and so represent the elements of So and g, respectively.

The S output of the array is from the right hand side, as shown in figure 5. As with the input, each element of this output is staggered by one cycle per element.

This process is repeated for each spatial constraint vector Ox (steps 405, 406), where Cx is the spatial constraint vector associated with the xth wanted signal, and the resultant outputs of the first pass are therefore a number of vectors Sx, one for each wanted signal.

The second pass through the systolic array involves inputting the vectors Sx each with its row order reversed, represented as gx reversed as shown in Figure 10, (steps 408-413). Prior to this operation, the stored values of the array cells are transposed (step 407), by performing a reflection around the non-leading diagonal. Strictly speaking, the transposition of the cell values transgresses the definition of a systolic array operation, but nonetheless such a procedure is simple to carry out within an FPGA, (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit), or other processing device. The resulting vector outputs Ox, again one output per wanted signal

(corresponding to input vectors Sx reversed), are a scaled version of the weights for the narrowband solution (RIMS) as shown in figure 10 for each wanted signal. - 14

The scaling factors r''2 are equal to the values of the boundary cells and hence are known which enables the weights to be calculated.

A shortcut in the weight calculation can be made at this point which avoids the need to scale the output by r''2 to obtain the narrowband weights. Inspection of the boundary cell equations given above reveals that the g output is equal to the xn input divided by r,,2. So the desired result can be extracted from the input of the boundary cells, without any need to scale the values.

Once the scaling factors have been dealt with by either of the above techniques the narrowband weight vectors Wx can be output (step 414) for use in a beamformer In some circumstances it may be possible to store the received signals whilst the weights are calculated and apply the weight vectors to the actual received signals used to calculate the particular weights, alternatively weights can be used to extract data which arrives after the calculation has been completed. If the receiving apparatus is moving slowly, however, it may not be necessary to calculate the weights constantly, but only periodically and a particular weight solution can be used for the period of time between calculations. If the jamming sources are also moving, this will additionally affect the performance as the time since the last weight calculation increases and necessitate more frequent calculations.

An example of a broadband receiving apparatus 1100 is shown in figures 11 and 12, which correspond to figures 2 and 3 in the narrowband discussion. Where appropriate, common reference numerals have been used. The operation of the receiving apparatus is discussed, by way of example only, for the situation where there are two wanted signals from two different satellites (as shown in figure 12).

A signal is received on each of four antennas 201 resulting in 4 streams of data X, - X4. This data is in digital form, having been demodulated and sampled by a receiver 207. Each signal received includes a combination of wanted signals, noise and interference. The four signals, and therefore data streams, will not be identical as each wanted or interfering transmission arriving at the array will have a different spatial signature (the relative phasings of the signals across the antenna array being a function of the direction of arrival).

A difference from the narrowband apparatus described previously is that in the broadband apparatus, the signal from each antenna is connected via a series of - 15 delay elements 1101. A representation of the signal at each delay tap 1102 is directed to the beamformer 202. Figure 11 shows 3 delay taps 1102 per antenna, each delay tap separated from the adjacent delay tap by a delay element 1101, there being 2 delay elements. Furthermore, a representation of the signal at each delay tap for each antenna is directed to a processor 204 via a tap 203. In this example, the processor thus receives 12 data streams, made up of 3 delay taps for each of 4 antennas. As with the narrowband case, the processor processes the data to determine the way in which the streams should be combined in the beamformer in order to extract the two wanted signals from to the noise and interference, the difference being that the weight vector will now have 12 elements, as opposed to 4 for the narrowband case. As described earlier, the processor may require information relating to the location of the sources of the wanted signals and this information may be obtained from a memory or database 205, or through the use of known techniques for determining the direction of arrival, or spatial signature, of a transmitting source.

In addition to this spatial constraint information, the processor will also require a vector of constraint gain factors for each time-domain delay tap to allow determination of the broadband weight vectors. The processor outputs two weight vectors W., W2, one for each wanted signal 301, 302, which are fed 206 to the beamformer. The beamformer then uses We to combine the received data streams to extract the first wanted signal and W2 to combine the received data streams to extract the second wanted signal. As described above, these weights We, W2, will contain more elements in the broadband case compared to the narrowband case, because the number of elements within the weights equals the number of antenna elements multiplied by the number of taps to the beamformer and there is only one tap in the narrowband case (leading to 4 x 1 = 4 elements) and 3 delay taps in the broadband case (leading to 4 x 3 = 12 in the example shown).

It should be understood that the situations described are by way of example only and there may be any number of taps and antenna elements depending on the required application. The adaptive performance improves with the number of delay taps, however this increases the processing load considerably, and there will be an optimum number of taps for a system based on this trade-off. For a system requiring around 60dB of jammer cancellation, the optimum number may be around six taps. - 16

The description below describes the processing activity within processor 204 in the broadband case, in more detail, with reference to the process steps shown in the flow diagram of figures 13-15. Where the processing steps are common between the narrowband and broadband cases, common reference numerals have been used.

The processor array used for the broadband situation is exactly the same as used in the narrowband case and described above with reference to figures 5-7 and the first part of the process is identical to that used in the narrowband case, steps 401-413, as shown clearly in figure 13. However, the resulting outputs Ox, to which are a scaled version of ROBS, form only the first part of the optimum weight solution for the broadband case, as shown below.

Using a space-time approach, with a tap-delay beamformer, as shown in figure 11, the optimum solution in terms of constrained least mean-squares has been derived by Frost ("An algorithm for linearly constrained adaptive array processing", Proceedings of the IEEE, Vol.60, no.8, pp.926-935, August 1972).

Minimisation of the array output power 'XHXW, subject to the wanted signal constraint CHW = E, leads to the optimum weight set given by where Wopt iS an elongated column vector representation of the optimum weight vector, M = XHX is the data covariance matrix, C is a block matrix constructed from spatial constraint vectors, and is a vector of constraint gain factors for each time-domain delay tap.

The constraint matrix C is built up from repeated copies of the narrowband constraint [c, c2 c3]T, which is the matched filter to the wanted signal at the centre of the frequency band. In a standard implementation, where a flat frequency response is desired, the time domain vector E has one element set to unit gain, typically the central element, with the rest to zero, as shown in figure 16.

The look direction constraint given in equation 1 is expressed in terms of the data covariance matrix M. However, in this case OR decomposition is used as the adaptive algorithm, in which case the covariance matrix is not explicitly formed. - 17

This requires an alternative constraint, expressed in terms of the decomposition matrices Q and R. The OR decomposition of the data matrix X produces a unitary matrix Q and an upper-triangular matrix R X = Q R (2) Thus the covariance matrix M is equivalent to M = XH X = (Q R)H Q R = RH (QHQ) R = RH R (3) since QHQ = I, the identity matrix. Substituting into equation 1 gives WOp, = (RHR)'C (CH(RHR)-' C)-' E (4) Defining S = R-H-C gives Having successfully calculated the first part of the optimum weight solution, namely R-'S, in the first two passes through the array, a further two passes are required in order to calculate the second part of the equation (SH S)-' 5.

By re-writing the term R-' S as its equivalent (XH X)'C, the similarity to this second part of the optimum weight solution becomes clear, with X being replaced by S. and the matrix C being replaced by the vector E. A slight complication arises in that matrix S is not known - the fact that a square-root free algorithm is used means that a scaled version, S. is obtained in its place. However, this issue can be overcome through the use of the scaling factor a contained within the boundary cell equations. This scaling factor set to '1' for all other array operations. Only in this third pass of the constraint is it employed fully.

To calculate the second part of the optimum weight solution, the systolic array is re-'nitialised (step 1401), with all stored values set to zero, and the matrix S. is 2s input to the array (step 1401). The array is allowed to adapt (Lm set to one), and the scaling factor is set such that a(k) = rkk2, on the kth cycle. The internal cell values are labelled ry Once S. has been input, the array is frozen, and the complex conjugate of the time domain constraint vector F is fed into the array (steps 1403 and 1404). The corresponding output vector, labelled By, is what is used in the final pass through the array. The process Is illustrated in figure 17. It - 18 is noted that on this pass, the systolic array will be much smaller- in the arbitrary example shown in figure 17, there are now just two columns. In actual fact, the array is reduced by a factor equal to the number of elements, as the new size is equal to the number of delay taps, (3 in the example shown in figure 11). For example, for a system with 7 antenna elements and 6 delay taps per element, the main systolic array will have 42 columns. For the third and fourth passes of the constraint, the array size is reduced to just 6 columns.

To complete the calculation of the second part of the optimum weight equation, a fourth and final pass through the systolic array is required. Firstly, the stored lo values rid are transposed (step 1405) in a similar manner to that used in the second pass (step 407). Then, the output vector from the third pass, Ad, has its elements reversed (step 1406) before being input to the array (steps 1407 and 1408). The process is illustrated in figure 18. Again, we can use the shortcut described above to extract the unscaled version of the desired output.

Having calculated both parts of the optimum weight solution for the first wanted signal (O. and Or'), a simple multiplication (step 1409) produces the required weight vector We.

In order to calculate the optimum weight solution for the second wanted signal, it is necessary to repeat these third and fourth passes (steps 1401-1409) but this time using matrix 2 as the input to the array (steps 1501-1509, resulting in output O2'). This enables calculation of the second required weight vector W2(by multiplying O2 and O2').The spatial constraint matrix Ox, as described above, enables the processor to attempt to steer a beam in the desired look direction i.e. towards the source satellite for each signal. Figure 19 shows some results from a simulation of the invention Figure 19 shows a typical adapted beam pattern produced by the simulator, corresponding to a single jammer scenario, with a beam directed at the wanted signal 1901 and a null steered towards the jammer 1902.

The description above assumes that the receiving apparatus knows the direction of the source of the wanted signal, which will be the case where the receiving apparatus has knowledge of its own position and orientation, an accurate timebase, and has access to satellite almanac information. If this position and time information is lost or not known, or the look up information relating to satellite position is lost, the relative location of the satellites may not be known. - 19

ln this situation, the receiving apparatus can adopt an alternative mode of operation, by selecting one of a number (e.g seven) fixed look directions. This is not a selection between fixed beams and the adaptive processor still acts to cancel jamming, rather it is the use of a number of fixed look direction constraints CX, which are stored in the receiving apparatus. Depending on the antenna array architecture, the fixed look direction constraints may be equally spaced around the receiving apparatus.

An example arrangement of the fixed look direction constraints is plotted in figure 20, which shows the unadapted combined effective beam pattern. A central lO beam points straight upwards, with six equally spaced beams directed at 45 elevation. To generate this plot, at every signal position in azimuth and elevation, the constraint which provides maximum output power (averaged across the frequency band) has been selected.

Preferably, the number of antennas on the receiving apparatus exceeds (Nammers + 1) where N,ammers is the number of jamming signal sources. This means that the adaptive process has a spare degree of freedom with which to steer a beam towards the wanted signal Where the number of antennas equals Nammers + 1, all the degrees of freedom are used in the cancellation of interference so there is no ability to steer a beam.

The process or algorithm described above may be implemented on an FPGA or ASIC, and being a systolic array implementation, a number of significant advantages are obtained. Being parallel in nature, the algorithm maxmises processing speed through capitalising on the parallel architecture of the FPGA or ASIC. In addition, the fact that the array consists of repeated blocks of operations means that any high-level design code, such as VHDL (Very high speed integrated circuits Hardware Description Language) code is both simpler to construct and fully scalable In terms of the number of beamformer elements.

The processing technique described above applies the look direction constraint after adaptation to the received data. This provides considerable advantages over techniques which apply the wanted signal constraint as a matrix transform before the systolic array, as this requires a separate systolic array for each wanted signal, which is clearly an unacceptable overhead. -

The processing technique described above has further advantages because by adjusting the boundary-cell scaling factors within the systolic array, it is possible to form a solution that does not require square roots. This factor is a significant advantage in FPGA implementations, where floating-point square root calculators are relatively large and slow.

Although the above description relates to detection of GPS signals from satellites, the Invention is not limited to this and is also applicable to other radio systems where it is necessary to extract more than one wanted signal from a set of received signals.

lo It will be understood that the above description of a preferred embodiment is given by way of example only and that various modifications may be made by those skilled in the art without departing from the spirit and scope of the invention. - 21

Claims

1. A method of calculating weights for use in extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, said method comprising the s steps of: feeding said plurality of received signals into a processor array; manipulating said array according to a plurality of spatial constraints, producing a plurality of output matrices, each one of said plurality of output matrices corresponding to one of said plurality of spatial constraints; lo reorientating said array and each one of said plurality of output matrices, producing a reorientated array and a plurality of reorientated output matrices; feeding back each one of said plurality of reorientated output matrices through said reorientated array; and outputting a plurality of weights from said reorientated array, each one of said plurality of weights corresponding to one of said plurality of reorientated output matrices, wherein each of said plurality of weights corresponds to a wanted signal.

2. A method of calculating weights according to claim 1, wherein said step of feeding said received signals comprises: passing said plurality of received signals through a processor array in a first adaptive mode, said array comprising cells arranged to perform predetermined operations.

3. A method of calculating weights according to any of the preceding claims, wherein said each of said plurality of spatial constraints comprises a matrix, and said step of manipulating comprises the steps of: passing sequentially each of said plurality of spatial constraints through the array in a second non-adaptive mode; and storing a plurality of output matrices, each of said plurality of output matrices corresponding to the passing of one of said plurality of spatial constraints. - 22

4. A method of calculating weights according to any of the preceding claims, wherein said step of reorientating comprises the steps of: reversing the order of the rows in each one of said plurality of output matrices, producing a plurality of reorientated output matrices; and transposing cells in said array, producing a reorientated array.

5. A method of calculating weights according to any of the preceding claims, wherein said step of feeding back comprises the step of: passing sequentially each one of said plurality of reorientated output matrices through said reorientated array in a second non-adaptive mode.

lo

6. A method of calculating weights according to any of the preceding claims, wherein each one of said plurality of spatial constraints is associated with a wanted signal.

7. A method of calculating weights according to any of the preceding claims, wherein each one of said plurality of spatial constraints is the Hermitian conjugate of a spatial constraint matrix associated with one of said wanted signals.

8. A method of calculating weights according to any of the preceding claims, further comprising the steps of: A. feeding a first of said plurality of output matrices into said processor array; B. manipulating said array according to a temporal constraint, producing a first output matrix; C. reorientating said array and said first output matrix, producing a second reorientated array and a first reorientated output matrix; D. feeding back said first reorientated output matrix through said second reorientated array, producing a first output vector of a plurality of output vectors; repeating steps A - D for each of said plurality of output matrices, producing said plurality of output vectors, each one of said plurality of output vectors corresponding to a wanted signal; and 23 multiplying each of said plurality of weights by the one of said plurality of output vectors corresponding to a same particular wanted signal, to produce a plurality of broadband weights, wherein each of said plurality of broadband weights corresponds to a wanted signal.

9. A method of calculating weights according to claim 8 wherein step A comprises: resetting said processor array to produce an initialised array; and passing a first of said plurality of output matrices through said initialised processor array in a first adaptive mode.

lo

10. A method of calculating weights according to any of claims 8 and 9 wherein step B comprises: passing said temporal constraint through the array in a second non-adaptive mode; and storing a first output matrix.

IS

11. A method of calculating weights according to any of claims 8-10 wherein step C comprises: reversing the order of rows in said first output matrix to produce a first reorientated output matrix; and transposing cells in said array, producing a second reorientated array.

12 A method of calculating weights according to any of claims 8-11 wherein step D comprises: passing said first reorientated output matrix through said second reorientated array in a second non-adaptive mode; and producing a first output vector of a plurality of output vectors.

2s

13 A method of extracting a plurality of narrowband wanted signals from a plurality of received signals comprising the steps of: calculating weights according to any of claims 1-7; passing said plurality of weights to a beamformer; and - 24 combining said plurality of received signals according to said weights in said beamformer to extract said plurality of narrowband wanted signals.

14. A method of extracting a plurality of broadband wanted signals from a plurality of received signals comprising the steps of: calculating weights according to any of claims 8-12; passing said plurality of broadband weights to a beamformer; and combining said plurality of received signals according to said broadband weights in said beamformer to extract said plurality of broadband wanted signals.

15. A method of receiving a plurality of wanted signals comprising the steps of: to receiving a plurality of received signals using a plurality of antenna elements; and extracting said plurality of wanted signals according to any of claims 13 and 14.

16. A processor for calculating weights for use in extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, said processor comprising: an input for receiving said plurality of received signals; a processor array; means for passing said plurality of received signals into said array; means for manipulating said array according to a plurality of spatial constraints; means for storing plurality of output matrices resulting from said manipulation, each one of said plurality of output matrices corresponding to one of said plurality of spatial constraints; means for reorientating said array and each one of said plurality of output matrices, producing a reorientated array and a plurality of reorientated output matrices; means for feeding back each one of said plurality of reorientated output matrices through said reorientated array; and - 25 an output for outputting a plurality of weights from said reorientated array, each one of said plurality of weights corresponding to one of said plurality of reorientated output matrices, wherein each of said plurality of weights corresponds to a wanted signal.

s

17. A receiver apparatus for extracting wanted signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of said wanted signals, comprising: a plurality of antenna elements, each said antenna element for receiving one of said plurality of received signals; in a processor according to claim 16; and a beamformer for combining said plurality of received signals according to said plurality of weights.

18. A radio communications network comprising a receiver apparatus according to claim 17.

19. An adaptive antenna apparatus for receiving a plurality of GPS signals comprising a receiver apparatus according to claim 17.

20. A method of extracting GPS signals from a plurality of received signals, each one of said plurality of received signals comprising a combination of GPS signals and a jamming signal, said method comprising the steps of: calculating weights according to any of claims 1-12; passing said plurality of weights to a beamformer; and combining said plurality of received signals according to said weights in said beamformer to extract said plurality of GPS signals.

21. A method of extracting GPS signals according to claim 20, wherein each one of said plurality of received signals comprises a combination of GPS signals and a plurality of Jamming signals.