Interference suppression in radio receivers
BACKGROUND OF THE INVENTION
Field of invention
The invention relates in general to interference suppression in digital radio receivers. In particular, the invention relates to co-channel and adjacent channel interference suppression in digital radio receivers.
Related art
A communication system can be seen as a facility that enables communication between two or more entities such as user equipment and/or other nodes associated with the system. The communication may comprise, for example, communication of voice, data, multimedia and so on. The communication system may be circuit switched or packet switched. The communication system may be configured to provide wireless communication.
The term cellular communication system refers to a system, where the coverage is provided by a plurality of cells. Communication devices communicate via a cellular communication system using one or more cells at a time. A communications device moving within the area of a cellular communication system typically changes cells depending on the quality of signals relating to the cells.
Frequency re-use refers to the use of a certain frequency band in a nearby cell. Typically the aim is not to use the same frequency band in neighboring cells. In general, at least two cells using other frequency bands are between two cells using a same frequency band. Due to, for example, limited number of available frequencies it may be necessary to use a given frequency band in cell which are near each other, even neighbors to each other. Use of the same frequency band in a nearby cell (or, in general, by any radio transmitter) typically causes co-channel interference to users using this frequency band in a given cell. Use of adjacent frequency bands in nearby cells causes adjacent channel interference.
In wireless communication systems the performance limiting factor is interference, rather than noise. Therefore, the capacity of the wireless communciation systems can be increased by the introduction of receivers with improved performance in interference limited scenarios.
One way of obtaining co-channel and adjacent channel interference rejection in wireless communication systems is to use an antenna array taking advantage of spatial diversity. This technique is, however, generally not feasible in portable communications devices due to cost, complexity and size constraints.
As an alternative to multiple antennas or antenna arrays, Single Antenna Interference Cancellation (SAIC) techniques have been studied. SAIC techniques can considerably improve receiver performance with minimum software upgrade in a communications device. One SAIC approach is to consider interference as colored noise. By whitening the colored noise, interference suppression and signal gain can be achieved.
The schematic structure of a conventional receiver is shown in Figure 1. The received signal Rx(t) 11 is first filtered with a band-pass filter (Rx-Filter 21), its output r(t) 12 is synchronized and de-rotated in a synchronization block (Sync 22). The synchronized and de-rotated signal x(t) 13 is then used for channel and interference estimation in a channel estimator (Ch-Est 23). The channel estimator provides at least a channel estimate h 14. At least the synchronized and de-rotated signal x(t) 13 and the channel estimate h 14 are provided to the equalizer (Equalizer 24). The equalized signal z(t) 15 is input to a decoder ( Decoder 25) for decoding, and the decoder 25 gives the transmitted bit s'(t) 16. The equalizer may be replaced with another type of detector.
In WO 0193439 the interference is modeled as an HR (infinite impulse response) process, and consequently the whitening operation is performed by a (multidimensional) FIR (finite impulse response) filter. WOO 193439 uses in-phase and quadrature components sampling, so that for each symbol there is one in-phase component sample and one quadrature component sample. The whitening operation in WOO 193439 is performed on the in-phase and quadrature components of the de- rotated signal x(t).
One problem associated with the whitening discussed in WOO 193439 is that the whitening operation using the FIR filter takes into account only correlation of in-phase and quadrature components between symbols. There may remain residual intra- symbol correlation where with intra-symbol we refer to all the samples falling within one symbol duration.
A further problem with the whitening discussed in WOOl 93439 is that if the signal noise is white, i.e., we are operating in a sensitivity limited scenario, the whitening using a FIR filter may colour the already white noise, leading to degraded performance
compared with a conventional receiver. The degraded performance is typically due to inaccuracies in the estimation of the FIR filter coefficients.
The present invention aims to address at least some of the problems discussed above.
SUMMARY OF THE INVENTION
In accordance with a first aspect of the invention, there is provided a method for suppressing interference, the method comprising providing a succession of samples relating to a succession of symbols, there being provided at least two samples per symbol, determining filter coefficients for a first whitening filter relating to suppression of intersymbol interference, whitening the succession of samples using the first whitening filter and the filter coefficients, and whitening the succession of samples using a second whitening filter, said second whitening relating to suppression of intrasymbol interference based on correlation properties of samples within a symbol.
In accordance with a second aspect of the invention, there is provided a computer readable medium containing executable computer program instructions which, when executed by a data processing system, cause said data processing system to perform a method as defined by the first aspect of the invention.
In accordance with a third aspect of the invention, there is provided a device for suppressing interference, the device comprising means for receiving samples relating to symbols, there being provided at least two samples per symbol, and whitening means for whitening the received samples, said whitening means having a first whitening filter relating to suppression of intersymbol interference, means for determining filter coefficients for the first whitening filter, and a second whitening filter relating to suppression of intrasymbol interference based on correlation properties of samples within a symbol.
There is further provided a communications device comprising means for receiving symbols over a radio interface,
means for providing samples relating to the received symbols, and a device in accordance with the third aspect of the invention.
There is also provided a network element for a communication system, the network element comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the third aspect of the invention.
In accordance with a fourth aspect of the invention, there is provided a method for suppressing interference, the method comprising providing a succession of samples relating to a succession of symbols, there being provided at least one sample per symbol, determining filter coefficients for a whitening filter based on a model jointly estimating filter coefficients and channel estimates, the whitening filter relating to suppression of intersymbol interference, whitening the succession of samples using the whitening filter and the determined filter coefficients, thereby providing a succession of whitened samples, and determining channel estimates based on the succession of whitened samples.
In accordance with a fifth aspect of the invention, there is provided a computer readable medium containing executable computer program instructions which, when executed by a data processing system, cause said data processing system to perform a method in accordance with the fourth aspect of the invention.
In accordance with a sixth aspect of the invention, there is provided a device for suppressing interference, the device comprising means for receiving samples relating to symbols, there being provided at least one sample per symbol, a whitening filter relating to suppression of intersymbol interference, a joint estimator for determining filter coefficients and channel estimates, the joint estimator configured to input the filter coefficients to the whitening filter, and a channel estimator for determining channel estimates corresponding to the whitened samples output from the whitening filter.
There is further provided a communications device comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the sixth aspect of the invention.
There is also provided a network element for a communication system, the network element comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the sixth aspect of the invention.
In accordance with a seventh aspect of the invention, there is provided a method for suppressing interference, the method comprising providing a succession of samples relating to a succession of symbols, there being provided at least two samples per symbol, determining filter coefficients for a whitening filter relating to suppression of intersymbol interference, and whitening the succession of samples using the whitening filter and the filter coefficients, wherein determining said filter coefficients for the whitening filter and whitening the succession of samples are carried out on the succession of samples in a fractionally spaced manner.
In accordance with a eighth aspect of the invention, there is provided a method for suppressing interference, the method comprising providing a succession of samples relating to a succession of symbols, there being provided at least two samples per symbol, dividing the succession of samples into a set of symbol-spaced successions of samples, determining filter coefficients for a whitening filter relating to suppression of intersymbol interference, and whitening the succession of samples using the whitening filter and the determined filter coefficients, wherein determining said filter coefficients for the whitening filter and whitening the succession of samples are carried out in parallel on the set of symbol-spaced successions of samples.
In accordance with a ninth aspect of the invention, there is provided a computer readable medium containing executable computer program instructions which, when executed by a data processing system, cause said data processing system to perform a method in accordance with the seventh aspect or the eighth aspect of the invention.
In accordance with a tenth aspect of the invention, there is provided a device for suppressing interference, the device comprising means for receiving samples relating to symbols, there being provided at least two samples per symbol, a whitening filter relating to suppression of intersymbol interference, and means for determining filter coefficients for the whitening filter, wherein said means for determining filter coefficients and said whitening filter are configured to process the succession of samples in a fractionally spaced manner.
In accordance with a eleventh aspect of the invention, there is provided a device for suppressing interference, comprising means for receiving samples relating to symbols, there being provided at least two samples per symbol, a whitening filter relating to suppression of intersymbol interference, and means for determining filter coefficients for the whitening filter, wherein the device is configured to divide the succession of samples into a set of symbol-spaced successions of samples, and the means for determining filter coefficients and the whitening filter are configured to operate using the set of symbol- spaced successions of samples.
There is further provided a communications device comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the tenth or the eleventh aspect of the invention.
There is also provided a network element for a communication system, the network element comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the tenth or the eleventh aspect of the invention.
In accordance with a twelfth aspect of the invention, there is provided a method for suppressing interference, the method comprising providing a succession of samples relating to a succession of symbols, there being provided at least one sample per symbol, determining filter coefficients for a whitening filter relating to suppression of intersymbol interference, and determining whether to whiten the succession of samples using the whitening filter and the determined filter coefficients based on the determined filter coefficients.
In accordance with a thirteenth aspect of the invention, there is provided a computer readable medium containing executable computer program instructions which, when executed by a data processing system, cause said data processing system to perform a method in accordance with the twelfth aspect of the invention.
In accordance with a fourteenth aspect of the invention, there is provided a device for suppressing interference, comprising means for receiving samples relating to symbols, there being provided at least one sample per symbol, a whitening filter relating to suppression of intersymbol interference, means for determining filter coefficients for the whitening filter, and switching means for switching the whitening filter into use based on said filter coefficients.
There is further provided a communications device comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the fourteenth aspect of the invention.
There is also provided a network element for a communication system, the network element comprising means for receiving symbols over a radio interface, means for providing samples relating to the received symbols, and a device in accordance with the fourteenth aspect of the invention.
BRIEF DESCRIPTION OF THE DRAWINGS
Embodiments of the present invention will now be described by way of example only with reference to the accompanying drawings, in which:
Figure 1 shows, as an example, a schematic block diagram of a conventional receiver;
Figure 2a shows, as an example, a schematic block diagram of a receiver in accordance with an embodiment of the invention;
Figure 2b shows, as an example, schematically a possible implementation of a channel estimation and signal whitening block in accordance with an embodiment of the invention in some detail;
Figure 2c shows, as an example, schematically a second possible implementation of a channel estimation and signal whitening block in accordance with an embodiment of the invention;
Figure 3a shows, as an example, schematically a channel estimation and signal whitening block having a joint estimator for filter coefficients and channel response;
Figure 3b shows, as an example, schematically a channel estimation and signal whitening block employing a conventional channel estimator; Figure 3c shows, as an example, schematically a channel estimation and signal whitening block employing a joint estimator for filter coefficients and a separate channel estimator for channel response;
Figure 4a shows, as an example, schematically a channel estimation and signal whitening block employing parallel symbol-spaced sample processing; Figure 4b shows, as an example, schematically combination of parallel symbol-spaced sample sequences using a combining filter;
Figures 5a and 5b show schematically handling of parallel symbol-spaced sample sequences;
Figure 6 shows, as an example, switching functionality relating to FIR whitening; Figure 7 shows, as an example, a filter module for removing intrasymbol correlations; and
Figure 8 shows, as an example, switching functionality relating to IQ whitening.
DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION
Embodiments of the invention are applicable in a digital communications system employing a modulation scheme which can be represented by a real modulation alphabet. Some examples of applicable modulation schemes are Pulse Amplitude Modulation (PAM), Minimum Shift Keying (MSK) modulation, Gaussian Minimum Shift Keying (GMSK) modulation, and Binary Phase Shift Keying (BPSK) modulation, and offset Quadrature Amplitude Modulation (offset-QAM) like binary offset QAM and quaternary-offset QAM, which can be viewed as binary or quaternary PAM signal by applying a proper rotation to every symbol. It is furthermore appreciated that those embodiments of the invention, which do not employ IQ-splitted signal, are applicable to any modulation, not only binary modulation.
Embodiments of the invention may be used, for example, in Global System for Mobile communications (GSM) or in Code Division Multiple Access (CDMA) systems. As embodiments of the invention employ a FIR whitening filter for suppressing interference, pilot symbols or other predetermined symbols (such as a training sequence) are generally needed for determining proper filter coefficients for the FIR whitening filter.
It is also appreciated that the description below assumes a single antenna receiver, and this is the situation in which the embodiments of the invention are most useful. The embodiments can, however, be easily extended to more than one receiver antenna, and the samples received from different antennas can be treated equivalently as fractional samples. For example, the algorithm would be the same if two samples per symbol are available from a single antenna or if one sample per symbol is available from two antennas.
Interference suppression is obtained by a digital processing of the signal, which can be classified as a filter or a succession of filtering operation on the digital signal with the aim to whiten the interference. In this description the term whitening filter refers to a filter or succession of filters.
A whitening filter in accordance with the described embodiments may combine the performance gain obtained by using a fractionally spaced processing and the performance gain obtained by splitting the received signal into its real and imaginary parts and processing this signal with a multidimensional filter. The real part of the signal is often referred to as the in-phase (I) component, and the imaginary part of the signal is often referred to as the quadrature (Q) component.
The schematic structure of a receiver 200, where whitening of the received signal is performed, is shown in Figure 2a. The receiver 200 includes a RX filter 210, a synchronising and de-rotating unit Sync 220, a channel estimation and signal whitening block 230, an equaliser 240 and a decoder 250. The RX filter 210, the Synch unit 220, and the decoder may be similar to those units in a conventional receiver shown in Figure 1. The equaliser 240 may be similar as the equaliser 24 in the conventional receiver shown in Figure 1. The equalizer may be replaced with another type of a detector.
The whitening operation in the channel estimation and signal whitening block 230 takes in most of the embodiments into account intersymbol correlation of noise and interference by modelling the noise and interference together as an autoregressive (AR) process and therefore assuming Infinite Impulse Reponse (HR) and intrasymbol correlation relating to noise and interference. The channel estimation and whitening block 230 thus typically contains a first whitening filter 231, which is a Finite Impulse Response (FIR) filter for removing intersymbol correlation of noise and interference, and a second whitening filter 232 for taking into account intrasymbol correlation. Typically the second whitening filter employs correlation information about multiple
samples related to the same symbol, for instance obtained by oversampling the signal or by considering as independent samples the in phase and quadrature signal component related to the same symbol, in other words, in-phase and quadrature signal components within a symbol. As a further option, signals originating from multiple receiver antennas may be used for suppressing intrasymbol correlation.
Regarding the output signals from the Channel estimation and signal whitening block 230, to a skilled person it is clear that an equalizer 240 needs a signal to be equalized and a channel estimate corresponding to the signal to be equalized. If the signal input to the equalizer is filtered, also the channel estimate needs to be filtered. The channel estimate corresponding to the filtered signal may be obtained, for example, by determining a channel estimate based on a signal before the filtering and then filtering this channel estimate using the same filter with which the signal is filtered. As an alternative example, the channel estimate corresponding to the filtered signal may be determined based on the filtered signal. Figures 2b and 2c shows two examples, but it is appreciated that these figures show the functionality typically present in the channel estimation and signal whitening block 230 and they are not intended to show the only possible arrangements of this functionality. For example, the second whitening filter 232 in Figures 2b and 2c (and also in Figures 3a, 3b, 3c and 4a) could contain functionality for estimating the channel h corresponding to the whitened signal y . In this case, there is no need for the separate channel estimator after the second whitening filter. As a second example, the second whitening filter may filter a channel estimate h ' input to the second whitening filter.
Figure 2b shows, as a first example, one possible implementation of the channel estimation and signal whitening block 230 in some more detail. The channel estimation and whitening block 230a in Figure 2b contains the first whitening filter 231 and the second whitening filter 232. The whitened signal y(t) output from the first whitening filter 231 is input to the second whitening filter 232. As the first whitening filter 231 is a FIR filter, the channel estimation and signal whitening block 230 contains functionality for determining filter coefficients for the FIR filter. In Figure 2b this is shown as a Filter coefficients block 233. Furthermore, the channel estimation and signal whitening block 230 contains functionality for estimating channel impulse responses. In Figure 2b, as an example, the block 230a contains a first channel estimator 234a, which takes the whitened signal y(t) as input and provides a channel estimate h ' corresponding to the whitened signal as output. The channel estimate h ' is input to the second whitening filter 232, which outputs a further whitened signal y . In
Figure 2b, there is a second channel estimator 234b for estimating a further channel estimate h corresponding to the further whitened signal y .
Figure 2c shows, as a second example, a second possible implementation of the channel estimation and signal whitening block 230 in some more detail. In Figure 2c, the block 230b contains a first channel estimator 234a before the first whitening filter 231. The channel estimate h output from the first channel estimator 234a is filtered with the first whitening filter, for obtaining the channel estimate h ' to be input to the second whitening filter 232. Similarly as in Figure 2b, the block 230b contains a second channel estimator 234b for estimating a further channel estimate h corresponding to the further whitened signal y output from the second whitening filter 232.
It is appreciated that although Figures 2b and 2c show the whitening using the first whitening filter 231 to occur before the whitening using the second whitening filter 232, the order of these two filters may be reversed. Furthermore, some embodiments of the invention may discard the second filtering block 232 and corresponding channel estimation. In this case, the filtered signal y(t) and the corresponding channel estimate h ' are typically input to the equalizer or other symbol detector.
As mentioned above, Figures 2b and 2c show the functionality typically present in the channel estimation and signal whitening block 230. It is not intended to show the only possible order of this functionality. For example, as is explained below, the channel estimation and filter coefficients may be determined jointly.
The filter coefficients A for the FIR filter in accordance with the above described model need to be determined for whitening the received signal. Figures 3a, 3b and 3c show some examples of determining the filter coefficients A.
In the first alternative Channel estimator and signal whitening block 330a shown in Figure 3a, there is a joint estimator 301 for determining the filter coefficients A and the whitened channel response K jointly. Details of jointly determining the filter coefficients A and the channel response h ' are discussed below. The filter coefficients A are input to the FIR whitening filter 231. The whitened signal y and the whitened channel response h' may be input to the second whitening filter 232 or, if the second whitening filter is omitted, to the equalizer 240.
In the second alternative Channel estimator and signal whitening block 330b shown in Figure 3b, the channel response h is first estimated in a first channel estimator 234a, which may be a conventional least square channel estimator or any other suitable channel estimator. Here it is assumed that the received transmitted symbol is a known
symbol, for example, a received transmitted pilot symbol. The channel estimator 234a provides a channel estimate h. Thereafter the received signal is reconstructed in the signal reconstructor 302 using the channel estimate h and the known symbol a. Using the received transmitted signal and the reconstructed signal, it is possible to provide, in the noise estimator 303, a noise estimate ή as the difference between the received
L-I transmitted signal and the reconstructed signal, e.g. hit) = x(t) -^ a(t - /)/?(/) . The
/=0 filter coefficients A may then be obtained as a function of the noise estimate ή in the filter coefficient estimator 304. The filter coefficients A are input to the FIR whitening filter 231. It is possible to determine the channel estimate corresponding to the filtered signal y using the channel estimate h and the filter coefficients A, for example, by filtering the channel estimate h with the FIR whitening filter 231, if the block 231 also receive the channel estimates h as input as Figure 2c shows. If a second filter 232 is present in the Channel estimator and signal whitening block 330b, the filtered channel estimate h ' may be input also to the second filter 232. Alternatively a second channel estimation for obtaining the channel estimate h ' is performed from the signal y(t). As a further alternative, the second whitening filter 232 may contain functionality for determining h ' based on h and A or based on the signal y(t).
A third alternative Channel estimator and signal whitening block 330c shown is shown in Figure 3c. Here, the joint estimator 301' is used for determining only the filter coefficients A. The filter coefficients are then provided to the FIR whitening filter block 231 and the signal is whitened. Thereafter, a first channel estimator 234a provides a channel estimate h' ' . The filtered signal y and the corresponding channel estimate h ' may be input to the equaliser. Alternatively, the block 330c may include a second whitening filter 232 and a second channel estimator 234b.
It is appreciated that although in most of the embodiments discussed in detail above include the second whitening filter for whitening intrasymbol correlation, this second whitening filter may be left out if it is, for example, expected that sufficient interference suppression can be obtained using only the FIR whitening filter. This may be the case, for example, when oversampling is used. A further reason for leaving the second whitening filter out is that the structure of the receiver can be kept simple. This is applicable especially for the structure shown in Figure 3 c, but may apply also for other structures of the channel estimation and signal whitening block 230.
The joint estimator 301 ' in Figure 3c may be the same joint estimator 301 as in Figure 3a. In this case, the whitened channel response K may be simply ignored. In the following it is discussed, how to determine the coefficients A and the whitened channel
K jointly. Then it is also discussed how to efficiently determine only the filter coefficients A in the joint estimator 301' without determining the whitened channel response.
Filter coefficients for the FIR whitening filter
In the following, by the way of example, details of the first whitening filter 231 are discussed, together with how to determine filter coefficients for the first whitening filter 231.
Modelling the noise and interference together as an autoregressive process and assuming Infinite Impulse Response (HR) is discussed next in more detail. In this model, the transmitted binary pilot symbols a(t), chosen from a binary alphabet are filtered by a complex- valued Finite Impulse Response (FIR) filter of length L (channel length). The desired signal received sequence x(t) relating to transmitted pilot symbols a(t) can be expressed as:
The signal may be split in real and imaginary parts a(f - I) Re{Λ(/)} (2)
X1 (t) = lm{x(t)} = Jm a(t - 1) Jm{h(ft} (3)
Introducing the following definitions: x(t){2.i} s ta(0 */(θr , X(tW+i),i} ≡ k(t) x'(t-l) -<t-K>r (4) ΛΛ(0) hR{\) ... hR(L -\) h{2,L) ≡ = [h(0) ... h(L-l)] (5) [A7(O) H1(I) ... hI(L-l)_ h(0) h(l) ... h(L -l) 0 0
0 h(0) h(l) ... h(L -l) 0 0
H {2(K+1),K+L} (6)
0 0 h(0) h(l) ... h(L -l) a(*W,i} s k0 <t-\) ... a(t-K-L + l)J (7)
the noiseless received signal can be expressed as:
X(t) = Ha(t) (8)
In this model, the co-channel interference and the white Gaussian noise are modeled together as an autoregressive process:
K n(t) = ∑ Amn(t -m) + e(t) with e(t) ~ WGN => E[e(t)e * (s)] = δt s (9) m=l
The above equation can be rewritten as
(10) and decomposed into the real and imaginary parts as
RQ{e(t)} = eR(t) = nR(t) ~ A^nR(t-l) + A(nI(t-l) + ...- AK RnR(t-K) + AK I n1(t -K)
(H)
Im{e(t)} = eI(t) = nI(t)-A(nR(t-\)-A?nI(t-l) + ...- AK 1nR(t-K)-AK RnI(t-K)
The matrix notation for the White Gaussian noise vector e(t) and the colored noise vector n(t) becomes:
e(t) = Wn(t) (12)
where it has been assumed real and imaginary part of n(f) are uncorrelated and the following definitions have been introduced:
1 0 A1 1 A1 2 ... Aκ x AK 2
W, R2(K+1)} o i - 4,3 - 4,4 - 4s:,3 - A (t)(2,n ≡ fc(0 ^(0]r(13)
K. e A n(t)mκ+m ≡ [nR(t) nj(t) nR(t-ϊ) n^t-V) ... nR(t-K) n,(t-K)f (14)
In accordance with this model, the noisy received signal corresponding to the pilot symbols is now:
X(t) = Ha(t)+ n(t). (15)
Multiplying all terms by W and reorganizing the equation, the following linear model is obtained:
WX(t) = WHa(t)+ Wn(t) = H'a(t) + e(t) => x(t) = Dz(t) + e(t) (16) where {\) ... h\ (K + L -l)
D 4,1 4,2 ••• 4Ϊ:,I 4Ϊ:,2 A1 H (0) h\
{2,3K+L} (17)
4,3 4,4 •" 4f,3 -^KA A1AO) Λ'j CO ... V1 (K + L-I)
z(tW,1} ≡ k(t-l) xT(t-2) ... xτ(t-K) | α(t) a(f-\) ... a(t-K-L + ϊ)J
and H'{2 A:+i} ≡ W^^^H^^ >K+L} is the channel filtered with the prediction error filter (i.e. the whitened channel). The length of the channel increases linearly with the filter order K. In this case the filter length is two times the length of the same filter where the splitting in real and imaginary part is not employed while the length of the filtered channel is the same.
Introducing the following definitions: X(t){ma] ≡ [x(t) x(t + l) ... x(t+R)f ,
E(t){β+U} - [e(t) e(t + l) ... e(t + R)f (18) and ^{R+lβK+L} ≡
xR(t -ϊ) X lit - K) a(t) a(t - K - L + \) xR(ή X lit) xjit-K + ϊ) a(t + \) a(t -K -L + 2)
xR(t + R - ϊ) xj(t + R -ϊ) ... xj(t - K + R) \ a(t + R) ... a(t + R - K - L + ϊ)
2K K+L
(19)
it is possible to write the linear system:
X(t)=Z(t)Dr +E(t) (20)
and solve it obtaining, for example, the least square estimation of the matrix D, containing the whitening filter coefficients and the filtered channel taps.
It is appreciated that although in the above model, the signal has been split into real and imaginary parts, the same model is applicable also without IQ splitting, for instance in the presence of multiple samples per symbols deriving from an oversampled signal or an interpolated signal or multiple reception of the signal by multiple antennas. By using also IQ splitting a more effective interference suppression in the case of binary modulation can be achieved, but if the signal is not IQ splitted the method can also be used for other than binary modulations.
As mentioned above, this modelling takes into account intersymbol correlation. This can be seen from Equation 13, where the two first columns of matrix W contain ones
and zeros. These two first matrix columns of matrix W relate to real and imaginary parts of the sample relating to the current symbol (or current time instant).
Regarding the third alternative shown in Figure 3 c, the FIR filter coefficients A can be determined without determining the whitened channel impulse response at the same time in the following way.
The solution of Equation (20) is given by OT {SK+L,2} = (zr (I)Z(I))"1 Zr(t)X(t) obtained considering the equivalent system
(zr(t)Z(t))Dr = Zr(t)X(t) (21)
In general, considering the particular structure of the terms in Equation 21, and introducing the definitions
a(t) ... a(t-K-L+ϊ) a(t + V) ... a(t-K-L + 2)
M ≡ (22) a(t + R) ... a(t + R-K-L + ϊ)
where a(t) are the known pilot symbols (or training sequence bits).
we can rewrite the solution of Equation 20 as
Applying the block matrix decomposition method Equation 24 can be rewritten as
[A = (xrEx)~1XrEr(t) E = I -M(M7M)-1M7
(25) |h'= (M7GM)-1M7 Gr(I) G = I -X(X71X)-1X7"
At this point we can solve only the first equation in (25) and obtain the whitening filter coefficients A, filter the signal x(t) and subsequently estimate the channel taps from the filtered signal x'(t).
It is important to notice that the matrix E does not contain terms varying from burst to burst and can therefore be pre-calculated.
The system in Equation 25 can be further simplified by applying the QR decomposition to the matrix E.
The linear equation system to be solved is
EXA = Er (26)
and applying the QR decomposition to the matrix E, Equation 26 becomes
QRXA = QRr (27)
but by definition Q1Q = I and therefore
RXA = Rr (28)
where now R is an upper triangular matrix with same rank as the matrix M.
In general R is only partially filled, and therefore some more processing resources can be saved considering its particular structure. Once again the matrix R is not varying form burst to burst and can be pre-calculated. The solutions of the equivalent system
(xrRrR X) A = X7R7R r (29) are the FIR filter coefficients.
Whitening sJRnals containing multiple samples per symbol
To obtain more information about the received signal, oversampling may be used. The signal received in the RX filter 210 is a digital signal, and front end filtering, down conversion and analog-to-digital conversion with oversampling is performed before the RX filter. Processing of oversampled data in the joint estimator 301, 301' and in the FIR whitening filter 231 is discussed next. Alternatively multiple samples per symbol may be obtained by interpolation, or they may be available because of multiple
replicas of the signal received by multiple antennas. In the following we will describe in more detail various options of dealing with multiple samples per signal where the number of available samples per symbol is denoted as NSPS.
Extending the previous formulation to the fractionally spaced domain may be done in various ways. A first option is to consider equation 1 where the discrete variable t span a fraction of the duration of the symbol period, and then following the already described steps. In this direct extension formulation the prediction error is given by
eW = Wrn(t)
= [l -A1 -A2 ... -A2K][n(t) n(t-V2) n(t-l) ... n(t-K)]T
The linear system we need to solve is still
X(t) =Z(t)Dr + E(t) (31)
where the following definitions have been introduced
Z(t){2R+2,6K+2L} ≡
χR(t - y2) X1Ct -V2) ... X1(T - K) a(t) 0
XR (t) X1 (t) X7 (Y- K + 1A) J 0 a(t- K - L + Y) (32)
xR(t + R) xj (t + R) ... xjjt - K + R + 'A) j 0 ... a(t + R - K - L + l) AK 2K+2L
x(t){2>i} ≡ h(0 *i(t)Y (34)
X(tW2j2} = [x(t) *(t + fc) - x(t + R + V4)F E(t){2Λ+2>2} ≡ [e(t) e(t + fc) ... e(t + R + 1Z2)]71
One thing to notice is that by doing so the first filter would already partially cancel also intrasymbol correlation, leaving to second filter only the job to cancel the remaining correlation within a fraction of symbol.
A possible and well performing solution is to split the oversampled data stream in NSPS (Number of Samples per Symbol) parallel streams, each symbol spaced, and then process the NSPS parallel data streams with NSPS parallel processing units, typically independently of each other. Referring to Figure 3 a, for example, this would means NSPS parallel joint channel estimation and filter coefficient estimators 301 and NSPS parallel first whitening filters 231.
This parallel symbol-spaced solution is shown schematically in Figure 4a, where a channel estimation and signal whitening block 430a is shown to contain a demultiplexer 401 for dividing the sequence of oversampled samples into NSPS sequences of symbol-spaced samples (in Figure 4a, as an example, into four sequences of symbol-spaced samples). Each symbol-spaced sample sequence is input to a respective joint channel and filter coefficient estimator 301. The filter coefficients A from these estimators are input to the respective first whitening filters 231. The NSPS parallel sequences of the whitened samples y are input to the second whitening filter 232.
It is appreciated that in Figure 4a the symbol-spaced sequences of samples are input to the second filter 232, because the detailed description below relating to the second filter 232 takes a sequence of fractionally spaced samples as input. Should the second filter 232 be implemented to take as input only one symbols spaced sample sequence, NSPS parallel second filters 232 may be provided in the channel estimation and signal whitening block 430.
The output from the (possible NSPS) second filter(s) 232 may be an oversampled whitened signal or, equivalently, NSPS streams of symbol-based sequences of whitened samples. This output may be sent to an equalizer that can handle an oversampled signal or NSPS parallel streams. Alternatively, the output from the second filter(s) 232 may be combined using a combining filter, which takes as input multiple parallel stream (or equivalently an oversampled signal) and gives a single stream symbol spaced as output. This combining filter may be done in several ways, for instance as a matched filter or the feedforward filter of an MMSE_DFE equalizer.
It is also possible that the second filter 232 contains functionality for combining oversampled signal into a sequence of symbol-based whitened samples.
Figures 5a and 5b shows schematically the demultiplexing and multiplexing discussed above. Figure 5b relates to an example, where NSPS=2.
As an alternative to simply multiplexing the NSPS parallel symbol-spaced sequences of samples, it is possible to combine the NSPS symbol spaced sequences of samples into one symbol-spaced sequence of samples using a matched filter. Figure 4b shows an example with a matched filter 403. It is possible that there is a second filter 232 between the parallel first whitening filters 301 and the matched filter 403.
A further solution is to use a formulation, which resembles in-phase and quadrature component processing. In this formulation the signal is "fractional spaced splitted" instead of being IQ splitted, that is the same processing that was followed on the IQ splitted signal can be performed on a signal where fractionally spaced samples are arranged similarly. In this formulation the prediction error is given by
e(t) = = Wn(t) = e(t)
(35)
1 0 - A 1,1 ~ AK,2 [n(t + V2) n(t) ... n(t - K)f
0 1 - A 1,3 - AA - ~ AK,4
Following the same reasoning as above, the linear system to be solved is
X(t)=Z(t)Dr +E(t) (36) where the following definitions have been introduced
x(t - lA) x(t - l) x(t - K) x(t + λA) x(t) x(t - K + \) Z(t) {R+l,3K+L) (37)
M x(t + R -lA) x(t + R -ϊ) ... x(t - K + R)
2K K+L
M is the same as in Equation (22)
Au \2 ... AKλ AKi2 \ h'CA) V(I + V2) W(K + L - 1A)
D {2,3K+L} (38)
1,3 AA αAT,3 %,4 h'(0) h\l) H(K + L -V)
x(t + lA) x(t) x(t + 3lA) x(t + Y)
X(t) {Λ+1,2} (39) x(t + R + 1A) x(t + R)
A further solution is to use the previous formulation together with the IQ splitting, therefore the error signal is defined as eR(t + V2) 1 0 0 0 -Ai ... -4,4 — "~4sr,4 nR (t + K) e, (t + 1A) 0 1 0 0 — A e(t) = = Wn(t) = -As ••• ~Αr,8 H1 (t + V2) eR(t) 0 0 1 0 ~ 4,9 ••• ~4,12 ••■ ~Α:,i2 eΛt) 0 0 0 1 ~ Al3 •" ~4,16 — ~ A,i6 (t- K)
(40)
Still, the linear system to be solved is
X(t)=Z(t)Dr+E(t) (41)
where the following definitions have been introduced
xΛ(f-fc) X7C-1Z2) X1V-K) xR(t + l/2) %j{t + V2) Xj(t-K + Ϊ)
Z(t) {72+1,5^+i} (42)
M xR(t + R-V2) X1Q + R-1A) ... Xj(t-K + R)\
AK K+L
M is the same as in Equation (22)
xR(t + V2) X1 (t + V2) xR(t) x}(t) xR(t + 3lA) X1 (t + 31A) xR(t + l) Xj(t + 1)
X(t) {«+1,4} (44)
_xR(t + R + V2) Xj(t + R + V2) xR(t + R) Xj(t + R)
In summary, multiple samples per symbol may be processed in series (that is, in one fractionally spaced sequence) for example in the order defined by one of the equations 30, 35 or 40 above. The multiple samples per symbol may, alternatively, be processed in connection with at least the FIR whitening by dividing the samples in symbol spaced streams and by processing the symbol spaced streams in parallel, typically independently of each other.
As can be seen from Equations 30 to 44, in fractional spaced processing methods the dimensions of the matrix Z(t) grow, whereas in the parallel symbol spaced method the matrix Z(t) is processed NSPS times. In fraction spaced processing methods, the size of the matrix Z (t)Z(t) which has to be inverted grows. Moreover, the more data columns are added, the more the matrix Z (t)Z(t) becomes close to singular and the inversion may become instable. This problem can of course be mitigated by adding to the matrix a small regularization term.
Switching FIR whitening on and off
Poor performance of a whitening receiver in a sensitivity limited scenario can be alleviated by assessing whether the noise is white or not and discarding signal whitening in the case of white noise. In the presence of white noise, the FIR filter coefficients A behave in a certain manner (possess values exhibiting certain characteristics). This manner can be determined, for example, by simulations. A decision to use FIR whitening may be based on the properties of the FIR filter coefficients A. In the following, a sensitivity detector refers to a block assessing the properties of the FIR filter coefficients A .
Figure 6 shows, as an example, a channel estimation and signal whitening block 630 using a FIR filter. As an example, the block 630 contains a joint channel and filter coefficient estimator 301, but any other alternative for determining the FIR filter coefficients may be used (see, for example, Figure 2b, 2c, 3b or 3c). The block 630 further contains a FIR whitening filter 231 and a sensitivity detector 601. The sensitivity detector 601 makes use of simple mathematical expressions in order to assess the FIR filter coefficients.
If the sensitivity detector 601 determines, based on the FIR filter coefficients, that the noise is more or less white, it switches in use the channel estimator 23. This way, when the noise is white and there is no need to remove co-channel or adjacent channel interference, whitening is not applied, but a (traditional) channel estimation is performed.
The decision about using the FIR filter may change from burst to burst. In a sensitivity scenario, whitening will rarely take place, that is, will take place in a small fraction of the bursts. In an interference scenario, whitening will take place very frequently, that is, in a large fraction of the bursts.
It is appreciated that determining whether to use FIR whitening based on the FIR filter coefficients can be used with any of the above described filters using a FIR whitening filter. It is appreciated that, as discussed in more detail below, also the second whitening filter relating to intrasymbol whitening may have associated functionality for switching on and off the use of the second whitening filter. Typically the switches relating to the first and second whitening filters operate independently of each other. If only one whitening filter has associated switching functionality, the other filter may be in use continuously. In some situations it may be feasible to have the switches operating in concert with each other. For example, the following co-operational cases may be feasible.
Case 1 : Whitening using the first and the second whitening filters is applied only when sensitivity detectors associated with both filters detect coloured noise (interference). If only one sensitivity detector (or none of them) detects coloured noise, no whitening is applied. Case 2: Whitening using the first and the second whitening filters is applied if either one (or both) of the sensitivity detectors detects coloured noise. Case 3: Only the sensitivity detector relating to the first whitening filter is present but it controls the use of the first and second whitening filters. Case 4: Only the sensitivity detector relating to the second whitening filter is present but it controls the use of the first and second whitening filters.
The sensitivity detector 601 applies at least one metric for assessing the FIR filter coefficients. A metric is a factor, whose value is dependent on FIR filter coefficients. As discussed below, a metric may be dependent on the complex numbers from pairs of FIR filter coefficients. It is possible to determine a range of metric values, which corresponds to white noise, for example by simulations. This range of metric values depends on the specific metric. If the sensitivity detector 601 judges that a metric value is within the range corresponding to the white noise, the FIR whitening is discarded. If a metric value is outside the range corresponding to the white noise, the FIR whitening is employed.
It is clear to a skilled person that the metric value range corresponding to white noise may be determined as a threshold which the metric should exceed or below which a metric value should remain, depending on the definition of the metric.
Referring now to Equation 13, the real-valued coefficients {A^} in the W matrix can be represented as complex numbers as follows (example with AR process of order K = 4):
Ci =ΛiΛ +jΛι>3, C2 =Ait2 +jAιA, C3 = A2,ι +jΛ2,3, ..., C8 = A4,2 +jA4ι4. In other words, complex numbers are formed from pairs of FIR filter coefficients.
A first specific example of metrics used in a sensitivity detector 601 is the following. Generally, metrics m and mn (n = 1 ... 2K— Y) can be formed as follows:
2K-I 2K-1 m = ∑Mn/,(|ReC,,+I|-|ReCΛ|)+ ∑vΛ/Λ(|lmCΛ+1| -|lmCΛ|) ;
«=1 n=l
Here, un and Vn {n = 1 ... 2K-V) are real- valued constants and function/, is defined for a given n {n = \ ... 2K-Y) as either fn(x) = x oxfn(x) = \x\.
In accordance with this first specific example, an interference scenario (coloured noise) is detected if the following assertion is true: m < mthr OK mi > U1 OR ... OR mL > aL where 1 ≤ L ≤ 2K-l.
If the assertion does not hold, a sensitivity limited scenario (white noise) is detected. The thresholds, mthr and αy ... ai depend on the signal levels in the system and must be tuned for a particular implementation. Likewise, the constants L, Un and Vn and the functions/, must be selected appropriately.
As an implementation example consider a GSM GMSK receiver with an oversampling rate of 2. The receiver jointly estimates a whitened channel estimate of length 6 (samples) and coefficients of an AR process of order K = 4 (samples). For each burst, after estimation of the coefficients {A^} in the W matrix, the following entities and metrics are computed:
CiR =AlΛ, C2 R =AK2, C3 R =A2Λ, C4 R =A2,2 Ci1= Au, C2 7=ΛM, Cj = A1Z, C4 !=A2A
Dx1 = IC1 7I, D2 1 = IC2 7I, D,1 = IC3 7I, D4 1 = \C/\
m = ~ Dl -D? -(Dξ -Dζ)+(ϋ* - D?)~ Dt1 -Dt - D[ -D[ -(D{ -D3')
W1 = C1 = (C1Y +(C/)
An interference scenario (coloured noise) is detected if the following assertion is true: m < mthr OR mi > athr OR nt2 > at/,r- If the assertion does not hold, a sensitivity limited scenario (white noise) is detected. The thresholds have the following values: mthr — 1.5, athr = 3.0.
A second specific example of a metric m used in a sensitivity detector 601 is the following. The metric m is formed as the sum of the squared Euclidean distances between the actual FIR filter coefficients and reference coefficients Ck-
Here N is an integer constant between 1 and the AR process order multiplied with the oversampling factor, i.e. l<iV<2^ for 2 times oversampling. The reference coefficients Ck, l≤k≤N, are constant complex numbers.
In accordance with this specific second example, an interference scenario (coloured noise) is detected if the following assertion is true: m < nithr
If the assertion does not hold, a sensitivity limited scenario (white noise) is detected. The threshold mthr depends on the signal levels in the system and must be tuned for a particular implementation. Likewise, N and the c^s must be tailored to the implementation.
As an implementation example consider a similar GSM GMSK receiver as above. For each burst, after estimation of the AR coefficients {Ahk} in the W matrix, the following entities and metric are computed: + jA1Λ
An interference scenario (coloured noise) is detected if the following assertion is true: m < mthr- If the assertion does not hold, a sensitivity limited scenario (white noise) is detected. The complex c#'s are given as follows: c, = 1.3e-jW4, c2 = 1.3e^/4, c3 = 12ejπ/2 = jl.2, C4 = 1.2^* = -1.2
Threshold value: mthr = -1.0.
Sensitivity switching using a sensitivity detector can significantly improve sensitivity performance of a receiver employing FIR whitening without hampering interference performance very much. Gains of up to 2.0 dB have been demonstrated for sensitivity performance in a realistic receiver for GSM phones. At the same time, the degradations in interference scenarios were limited to 0.3 dB for single interferer cases (co-channel interference or adjacent channel interference). Negligible degradations (<0.01 dB) were seen in a complex interference scenario with a mixture of several GMSK modulated co- and adjacent channel interference contributions. Such complex scenarios are the most realistic interference scenarios, and the losses in the single interferer cases are not important.
Intrasymbol whitening
In the following, handling of intrasymbol correlation in the second whitening filter 232 is discussed. Here is assumed that the second whitening filter 232 operates with two samples per symbol and IQ splitting. However, there may be any number of input samples per symbol, either because of oversampling or because of the presence of multiple receiver antennas. The IQ splitting is also optional. The number of samples per symbol can be increased also by means of interpolation if additional samples are not available directly from the received signal Rx(t). In general, the intrasymbol correlation can be handled by using correlation properties of samples relating to the same symbol.
For a skilled person it is evident how to apply the principles discussed below based on, for example, multiple samples per symbol obtained from IQ splitting, oversampling and/or multiple receiver antennas.
The following notation uses a signal y as input signal to the intrasymbol whitening filter. This is not intended to limit the signal processing to cases, where there is a FIR whitening filter before the intrasymbol whitening filter. As mentioned above, the order of these filters may be reversed.
Sampling of the received signal provides a sampled received signal y(nT) where T stands for the time between (transmission of) successive symbols (i.e. the inverse of the symbol rate), and n = k + j where k is the index that runs for all the transmitted symbols, q is the oversampling index that runs from 0 to / , and / is the oversampling factor, i.e. the number of samples per symbol,(or the number of receiver antennas or the oversampling factor x number of receiver antennas).
The second whitening filter 232 is a filter for filtering the sampled received signal to remove co-channel interference and noise so as to produce a filtered sampled signal γk (i.e. a succession of filtered signal samples). The input of the filter 232 is samples y(nT) of the (FIR whitened) signal and the corresponding estimated discrete channel impulse response h'(sT) , where s = m + q/l and m runs from 0 to v with O ≤ m ≤ v and v equal to one less than the channel impulse response length.
In the description that follows, over-sampling by an over-sampling factor of I=Q is assumed and the received signal samples yk at the input of the second whitening filter is written in terms of the channel data input gk and the sampled channel impulse response h'm as:
v J k = Ϋ Z-ih' m(0V &k0-)m + + Ϋ ZJ Ϋ ZJh' Ή°V ° kω-m + +n U (45) m=0 =1 »1=0 where O ≤ m ≤ v and v is equal to one less than the channel impulse response length. The superscript j denotes the indexing for each signal on the selected communication channel, with j =0 indicating the desired signal and the other j values indicates co- channel interference, and with
and using T for the time between (transmission of) successive symbols. We can define a block of Nf received samples as:
or more compactly, as:
M ω
Υk+Nf-ϊk ~ ** O -sfc+JVy-lt-v "** / , ■"■ i Sk r-+Wc-lfc-v + ^ n"A,r+Wy-lϋ: ' (46)
M which includes the desired signal Wog^+'N _u_vand the noise plus (co-channel) interference signal _u . For convenience, we define the noise
plus interference signal ik at the instant corresponding to index (sample counter) k as the 4x 1 vector:
Next, we define a filter operation L 1 so as to provide the whitened received signal samples y as follows:
where, according to the invention,
in which RH is the 4x4 noise plus interference correlation matrix given by:
in which E[...] is the mathematical operation of taking the ensemble average. Since the interference arises from a cyclo stationary random process, the expectation operation can be replaced by a time average, i.e.
Thus, each 4x 1 vector γk representing one symbol can be whitened as y* = Wy, (51)
where W is defined as the inverse of a square root operation on a positive definite Hermitian matrix R,, , i.e.
W = R-1/2. (51.1)
In some embodiments, the Hermitian matrix R,, may be the noise plus interference correlation matrix, and the whitening matrix W = R~1/2 may be obtained as the inverse of a Choleski factorization of RH . Alternatively the whitening matrix W can be obtained by any factorization of R,( , i.e. any factorization indicated by, (WW)-1 ^ R,, , (51.2) or be obtained by any factorization of R(( , as indicated by, UλV' = R L», ' such as for example by Singular Value Decomposition (SVD). In such a case, the
_y whitening matrix W can be obtained as W = V λ n .
According to some embodiments of the invention, a whitened impulse response H0 is also provided by the IQ filter, this time acting on the impulse response H'o :
Therefore, each 4 x 1 vector hm may be computed as:
for 0 < m < v . So, it is clear that the interference suppression method disclosed by the present invention does not cause the problem of increased channel length, i.e. it is clear from eq. (53) that hm has the same dimension as h'm .
Referring now to Figure 7, the filter module 732 may include: a first module 701 for constructing the succession of samples yk for k = l,...,Nf , with each k corresponding to one symbol; a next module 702 for determining the noise plus interference correlation matrix Rr, ; and a third module 703a for calculating the filtered, whitened samples yk = Wy4 for k = 1,...,Ny .
Referring still to Figure 7, in some embodiments the filter 732 can include an additional module following the module 703 for calculating yk = Wyt . Such a module would not decimate the signal in case of a fractionally spaced equalizer, but in case of a symbol-spaced equalizer, the module would decimate the signal, and would
do so by filtering the signal with a filter equivalent to the feed-forward filter of a DFE equalizer or by a polyphase matched filter.
The equalizer following the intrasymbol whitening filter 732 may be a trellis detector, and the input of the trellis detector is symbol spaced. This symbol spaced input can be obtained by post processing the output of the whitening filter W by a pre-filter (designed for white noise) equivalent to the feed forward filter of a Decision Feedback Equalizer (DFE). By approximating the noise plus interference correlation matrix R;( as in eq. (50), only the correlation within one symbol is removed. Therefore, some residual correlation may be expected, and could be, for instance, taken into account in a subsequent optimization of a feed-forward filter. After such post processing, a Forney metric can be used in the trellis Viterbi equalizer. See e.g. "Unification of MLSE Receivers and Extensions to Time Varying Channels"; by G. Bottomley, S. Chennakesku; in IEEE Trans. Inform. Theory; vol. 46, pp 464-472, April 1998.
Alternatively the equalizer could employ the Ungerboeck metric, also described in the above paper by Bottomley et al. In such a case, there is no need to employ a pre-filter, but the output of the whitening filter W can be decimated to one sample per symbol by employing a polyphase matched filter, matched to the whitening desired impulse response.
Alternatively the equalizer can operate with more than one sample per symbol, and in such a case there is yet another alternative implementation. If the equalizer can operate with more than one sample per symbol, instead of using the intrasymbol whitening filter , the same result can be obtained by modifying the metric inside the equalizer to take into account the noise correlation. More specifically a modified branch metric given by:
can be used instead of the Euclidean metric,
which would be used after the intrasymbol whitening operation.
Some embodiments of the invention also provide a mechanism for enabling and disabling IQ whitening (either performed as described above or using other techniques). Such a switching mechanism is useful because, it turns out, IQ whitening to suppress co-channel interference is remarkably effective for binary modulated co-
channel interference-limited channels, but causes some performance loss in sensitivity (white noise) limited channels. The invention thus provides a switching algorithm that can be used to dynamically switch between IQ whitening and non-whitening in a receiver. In some embodiments, the switching is based on examining relative values of different components of the noise plus interference correlation matrix Rr; 5 as explained below. In some other embodiments, the switching is based on examining expectations E[...] of different products i^ and ^i4+1 , as also explained below. The idea behind the switching mechanism provided by the invention is to determine whether the noise present is white noise (in such case the whitening is turned off), or non white noise, i.e. whether it is temporally correlated (in such case the whitening is turned on).
In the embodiment based on examining relative values of different components of the noise plus interference correlation matrix R;( , we detect whether the noise is white as follows. As described above, the IQ whitening filter (with truncated autocorrelation) requires the computation of the 4x4 (2-IQ split, 2-over-sampling) correlation matrix R/( from the noise samples. So in the preferred embodiment, the metric for switching on and off whitening is derived as a function of the elements of Rn .
In the case of binary modulated co-channel interferers, it turns out that the temporal correlation (between the fractional samples, i.e. the samples within one symbol) can be expressed as a combination of the elements of the correlation matrix R11 , as follows:
Rtempomr R11
*(RH (2,3)- Ra (1,4)). (56)
Thus for a binary modulated interfering signal, the temporal correlation can be obtained from the elements of the correlation matrix, and the correlation between the real and imaginary components of a sample does not become zero.
The switching provided by the invention according to the embodiment based on examining relative values of different components of the noise plus interference correlation matrix R;( can thus be prescribed as follows. Define
as a metric for indicating whether a channel is temporally correlated (i.e. is sensitivity limited so that the noise is white noise). As can be seen, the metric depends on the relative values of different components of the noise plus interference correlation matrix R,, . Then, if Mtc > τtc for some predetermined threshold τtc, the channel is
categorized as temporally correlated and whitening is performed; otherwise whitening is disabled.
As mentioned, in some other embodiments the switching is based on examining expectations E[...] of different products i^ifc and i^iA+1. In such embodiments, we evaluate the (numerical) value,
and also the (numerical) value,
Then, using as a metric for determining whether to whiten the quantity Mn defined by:
Mn = ^, (60) we switch to whitening when Mn is greater than some predetermined threshold τn
(having a value that can depend on the application).
Both of the embodiments for switching described above are based on observation of the second order statistics of the interference signal. Other embodiments may depend a more complicated measure.
Thus, referring to a filter block 800 shown in Figure 8, in the case of insubstantial temporal correlation (i.e. e.g. in the case of Mtc < τtc), the sensitivity detector 801 switches to the module 802 for filtering without whitening, and otherwise switches to the module 732 for whitening. In some embodiments of the invention, the module 802 may be discarded, especially if, for example, intersymbol whitening functionality is residing in the filter block 800 before the switch 801 or after the block 732.
In a filter, where there is a first whitening filter for suppressing intersymbol interference and a second whitening filter for suppressing intrasymbol interference, there may be a sensitivity detector and a switch (that is, switching functionality) relating to each filter. This has been discussed above in connection with switching on/off the intersymbol whitening filter.
The channel estimation and signal whitening block in accordance with embodiments of the invention may be implemented using software, hardware or a suitable combination of these. For example, embodiments of the invention may be implemented as an Application-Specific Integrated Circuit (ASIC), designed for a particular application. As a second example, embodiments of the invention may be implemented as program code for a programmable Digital Signal Processing (DSP) chip.
It is appreciated that interference suppression in accordance with embodiments of the inventions may be used in communications devices and in network elements of various communication systems. The terms communications device and network element refer here to any devices that are provided with functionality to receiving signals over a radio interface and processing the received signals. Some specific examples of the communications systems, where embodiments may be used, are cellular communications systems, such as GSM or UMTS (Universal Mobile Telecommunications System).
It is appreciated that although this description explains specific details of a FIR filter for whitening intersymbol interference and of a filter for whitening intrasymbol interference, it may be possible to implement the intersymbol and intrasymbol interference suppression using filters, which are different from the specific examples discussed above.
It is appreciated that, using the above mathematical formulation as an example, in the appended claims a succession of samples typically refers to x(t) for different values of t. Similarly, a succession of whitened samples typically refers to y(t) or to y (t) . Channel estimates typically refer to h, h ' or h , depending on the context.
It is appreciated that the handling of the equations and mathematical formulae discussed here in detail is not intended to be the only way of performing calculations in embodiments of the invention. It is clear to a skilled person that various modifications may be feasible.
It is appreciated that in the appended claims the wording "at least two samples per symbol" covers samples relating at least to oversampling (including interpolation), IQ splitting and/or multiple receiver antennas.
It is appreciated that in view of the foregoing discussion, it is clear that any feasible combinations of features, which are not contrary to the above description, are implicitly disclosed herein. More particularly, specific details relating to, for example, a certain block disclosed in connection with a specific embodiment are applicable to the same block when present in another embodiment, even if not expressly mentioned above.
Although preferred embodiments of the apparatus and method embodying the present invention have been illustrated in the accompanying drawings and described in the
foregoing detailed description, it will be understood that the invention is not limited to the embodiments disclosed, but is capable of numerous rearrangements, modifications and substitutions without departing from the spirit of the invention as set forth and defined by the following claims.