WO2002078205A1

WO2002078205A1 - Demodulating and despreading in cdma receivers

Info

Publication number: WO2002078205A1
Application number: PCT/GB2002/001438
Authority: WO
Inventors: Andrew Robert Nix; Yan Qint Bian
Original assignee: The University Of Bristol
Priority date: 2001-03-26
Filing date: 2002-03-26
Publication date: 2002-10-03
Also published as: GB0107612D0

Abstract

A receiver for receiving CDMA-type signals is arranged to select the most appropriate mode of reception for the received signals on the basis of one or more properties of the communications channel. The possible reception modes include a correlation mode, a rake mode and a LMMSE mode. The channel properties that are measured include the rms delay spread and the channel path power.

Description

DEMODULATING AND DESPREADING IN CDMA RECEIVERS

The invention relates to demodulating and despreading performed on received signals in a CDMA communications system.

The task of a CDMA receiver is to separate (via despreading) data belonging to the wanted user from data belonging to the interfering users. In a CDMA system, ignoring the effects of multipath, downlink (DL) interference is suppressed since the user codes are orthogonal and perfectly time aligned. However, in the uplink (UL), the user codes are sent over independent imperfectly time aligned channels and significant cross correlation interference occurs. Importantly, in the presence of multipath delay spread, UL and DL interference increases significantly due to the loss of code orthogonality. For this reason, independent UL (basestation receiver) and DL (mobile receiver) architectures are required. Interference from other users, known as Multiple Access Interference (MAI), is generally the dominant error mechanism and seriously restricts the capacity of a CDMA system.

For a single DS-CDMA user operating in simple Additive White Gaussian Noise (AWGN), a correlator represents the optimum receiver structure. A Rake structure represents the optimum DS-CDMA receiver structure for a single-user operating in a multipath delay spread environment [3]. For multi-user applications (as seen in practice), MAI seriously degrades system performance and neither of the above circuits are optimum [4]. Under these conditions, a circuit is required than can suppress the impact of MAI.

In a practical UMTS system, the reception of the desired user code suffers interference from the presence of other codes in the multipath propagation channel. The cross correlation noise of other users introduces a so-called interference floor, and the resulting system becomes interference limited, rather than noise-limited [5]. Unless suitable receiver processing is applied, the multipath channel will introduce MAI and thus limit the capacity and performance of a DS-CDMA system.

It is well known that Multi-User Detectors (MUD) offer significant improvements in DS-CDMA communications [4]. The optimal MUD takes the form of a Maximum Likelihood Sequence Estimator (MLSE). This structure offers the best performance in MAI, however its complexity increases exponentially with the number of users present. This complexity makes low cost, low power implementation extremely difficult in practice. Therefore, over the last decade, research has focused on sub-optimal MUD solutions offering feasible implementation [6]. Well known examples include the multipath-decorrelating receiver and the LMMSE (Linear Minimum Mean Square Error) receiver. These circuits suffer from poor performance at low signal to noise ratios and low rms delay spreads.

In the conventional multipath-decorrelating receiver, noise enhancement increases with the number of resolvable paths [7]. The adaptive LMMSE requires a symbol level training sequence to be sent from the desired user to the receiver. When applied to the UMTS TDD-CDMA standard, this is a major limitation since no such sequence is specified. To overcome this problem, blind LMMSE training is required. The basic non-reconfigurable LMMSE represents a well known circuit. Performance results for a fixed (in terms of Rake branches) LMMSE-Rake are provided in [9].

Due to natural time variations in both the mobile channel and the interference, satisfactory Bit Error Rate (BER) performance for a practical system cannot be obtained using a fixed-branch receiver structure.

A number of well known receiver architectures exist for DS-CDMA. These include the correlation receiver (optimum in a single user AWGN channel), the Rake receiver (optimum in a single user multipath channel) and the blind adaptive LMMSE receiver (sub-optimum in a multi-user multipath channel). In addition, an adaptive DFE (Decision Feedback Equaliser) chip-level equaliser can be used to restore user code orthogonality on the downlink of a CDMA transmission.

The circuit block diagram for an LMMSE receiver is shown in Figure 1. The adaptive weights are calculated blindly based on a data derived symbol sequence (see equation 5). The role of the LMMSE filter is to despread the desired CDMA signal on each rake finger and to adapt in a manner to reduce MAI. The LMMSE operates at the far lower symbol rate, unlike chip level equalisers that operate at the chip rate. The LMMSE is therefore a low complexity solution for the removal of MAI and the realisation of high capacity networks.

The problem with the well known LMMSE is its high sensitivity to background thermal noise. When performance is averaged over a time-varying delay spread channel, the applicant has found that, at high signal to noise ratios, the LMMSE operates well, however at low signal to noise ratio the structure breaks down [1][2].

An aim of the invention is to enhance the reception of signals transmitted using a CDMA scheme.

According to one aspect, the invention provides receiving apparatus for receiving CDMA-type signals, wherein the apparatus configures itself to a demodulating and despreading mode most suited for the signals received on a channel in accordance with measured properties of said channel.

The invention also consists in a method of receiving CDMA-type signals, wherein the reception mode is configured to a demodulating and despreading mode most suited for signals received on a channel in accordance with measured properties of said channel.

The invention also consists in a method of configuring receiving apparatus for receiving CDMA-type signals, wherein the apparatus has a number of available modes, and the method involves selecting the most appropriate demodulating and despreading mode for signals received on a channel in accordance with measured properties of said channel.

Hence, the invention improves the reception of signals by ensuring that the signals are being received using the most suitable demodulating and despreading mode chosen from a number of such modes that are available.

The mode used for the reception of the signals may be chosen on the basis of delay spread within the channel and/or a signal-to-noise ratio for the channel. These parameters are useful for allowing a mode selection to be made from amongst certain kinds of modes. The modes for which these channel parameters would make useful selection criteria include, for example, a correlation mode, a rake mode and a LMMSE mode. One or more parameters used for selecting a suitable mode may be obtained by performing a channel estimation process on the channel being received.

In a preferred embodiment, the available modes include one or more of a correlation mode, a rake mode and a LMMSE mode.

In one embodiment, a number of received components, which differ from one another by relative path delays, are used in the demodulation and despreading of the received signal. The components may be obtained by comparing the received signal to a threshold. Where several components exceed the threshold, the number of components used may be limited, for example by choosing a predetermined number of the strongest components. Where only one component is obtained, the demodulation and despreading mode selected becomes a correlation mode.

In one embodiment, the receiving apparatus according to the invention is based on a number of fingers (as might be used in a rake receiver), the signals on which are combined in the process of signal demodulation and despreading. The number of fingers used in the reception process can be set in dependence upon the delay spread of the received channel. Each of a number of the fingers may include an adaptive filter. At least one of the adaptive filters may perform a LMMSE process. Where the apparatus includes one or more adaptive filters, the adapation rates of the filters can be altered in accordance with the measured properties of the channel. Adaption rate adjustments can be made on the basis of a signal-to-noise ratio for the channel. It should be noted that it is possible to set the adaption rate of a filter to zero. Where an adaptive filter performs an LMMSE process, setting the adaption rate of the filter to zero means that the filter no longer performs an LMMSE process and the filter's operation becomes a normal despreading process.

According to another aspect, the invention provides apparatus for receiving CDMA-type signals, the apparatus being capable of despreading and demodulating signals received on a channel by using either a rake process or a LMMSE process, the apparatus being capable of selecting one of said processes for use with received signals on the basis of measured properties of the channel.

The invention also consists in a method of configuring a receiver for CDMA-type signals, the receiver being capable of despreading and demodulating signals received on a channel by using either a rake process or a LMMSE process, the method comprising selecting one of said processes for use with received signals on the basis of measured properties of the channel.

By way of example only an embodiment of the invention will now be described by reference to the accompanying drawings.

While any modulation scheme can be applied with this invention, for the purposes of illustration, BPSK is assumed in this particular embodiment. The modulation is sent over a frequency selective fading channel assuming K active users. An orthogonal Walsh code of length 16 is used in the spreading process (as defined in the UMTS TDD mode). The wideband channel distortion is modelled using a time variant tapped delay line transformation. The impulse response of the UL multipath channel for the k-t user can be represented as:

_k (t) = ∑ h_IΛδ(t - τ_ik )

'=' (1)

where h_k(t), L and r_/,* represent the complex channel gain, the number of resolvable paths and the excess delay respectively.

The channel model consists of several path clusters (resulting from multiple reflection and diffraction). The fading processes on each path are statistically independent and Rayleigh distributed [10]. It is assumed that the channel multipath delay spread, T„, & L/W, is comparable or greater than the chip period T_c, where W represents the signal bandwidth. Hence, the multipath will introduce ISI (Inter Symbol Interference) and the propagation channel is affected by frequency selective fading. For a DS-CDMA system, this ISI will cause serious MAI. The received signal can be mathematically expressed as:

r( =

where 2M+1 is the number of data symbols and Ak, bk, T_b and s_k denote, respectively, the amplitude, data sequence, bit period and signalling waveform of the k-t user. n(t) represents the complex zero-mean background Additive White Gaussian Noise (AWGN) with power spectral density σ². At the receiver, the signal r(t) is passed through a matched Root-Raised Cosine (RRC) filter with a roll-off constant,α, of 0.22. The output from this filter is then sampled at the chip rate.

For the UL, the received waveform is generated as the summation of the signals from each user, which is modelled as the convolution of the user signal with its independent channel. Hence, each user becomes a source of interference to all other users.

In contrast, for the DL, a common channel of the same form is assumed for all users. Ideally, there is no interference amongst users without the presence of multipath fading. However, in the presence of multipath, significant MAI will be observed. Consequently, to guarantee system performance in a CDMA system, the mitigation of ISI and MAI has to be considered together in an advanced receiver implementation.

A reconfigurable TDD mode detector is now developed to suppress MAI and ISI over noisy, frequency selective, time varying channels. More specifically, a reconfigurable blind adaptive LMMSE circuit is proposed with dynamic parameter settings and path selection. The architecture of this reconfigurable adaptive receiver is shown in Figure 2, where the LMMSE blocks take the form shown in Figure 1.

The reconfigurable circuit comprises a number of rake fingers, separated by time delays equal to a multiple of the chip period. In a practical embodiment, the number of rake fingers would be limited to reduce implementation complexity. Here, up to three rake fingers are dynamically assigned based on the results of the channel estimation process. Channel estimation is used to obtain information on the complex path weights (h_k). In addition, based on the power and time delay of the path estimates, approximate values for rms delay spread and Eb/No are generated (the latter using knowledge of the AGC setting and the receiver's noise floor). Using this information, the circuit is configured with the number of rake taps (L), the adaptive LMMSE step size, μws, and the matched filter co-efficient for each rake path.

As discussed, radio channel parameters such as rms delay spread and Eb No must be estimated to determine the value of the adaptive step-size, μ s If the rms delay spread value is low (below 75ns in the present embodiment), the value of μ_IMs is set to zero for all paths (i.e. the LMMSE adaptation circuit is placed in a power saving mode). This implies that there is no need to support the adaptive LMMSE (mode 1) and the resulting receiver is reconfigured as a conventional Rake structure (mode 2). The channel estimation data is analysed to dynamically determine the number of rake taps, L, required and their path delays. The L strongest path delays are chosen (see figure 3), where L represents the number of paths where the power exceeds a minimum threshold (set at -12dB relative to the peak in this particular embodiment). If I exceeds L_max, then the L_max strongest paths are chosen - figure 3(iii). If only one significant multipath delay branch is detected in the channel estimation process (i.e. L=\), then the architecture is effectively reconfigured as a correlation receiver (mode 3) - Figure 3(ii). For larger values of rms delay spread, μ_IjMS is set to an optimum value (determine via simulation for this particular embodiment) and blind adaptive LMMSE is performed. The adaption operates using decision directed (data-derived) training and aims to subtract MAI prior to multipath combining. MRC (Maximal Ratio Combining) and EGC (Equal Gain Combining) represent two well known algorithms that can be used to combine information from the rake fingers. Compared with EGC, MRC offers a greater diversity gain and is widely employed in the Rake combiner [11].

The operating modes of the reconfigurable receiver are summarised in Figure 4. The threshold value of Eb/No between mode 1 and mode 2 is determined by system simulation and is different for UL and DL operation. Similarly, the rms delay spread thresholds between modes 1, 2 and 3 are also determined via simulation. Optimum values for a UMTS TDD-CDMA embodiment are given in [1][2]. Table 1 summarises the values of the adaptive parameters in each mode. It should also be noted the path delay for each chosen finger is also an adaptive set of parameters that are passed to the receiving circuit.

Table 1: Adaptive Parameter settings as a function of mode

As summarised in Figure 3, the strongest branch path, together with other significant paths over a certain power threshold, are selected to determine the total number of branches (or taps) required for combining. The selected paths are continuously monitored to ensure that they do not fall below the relative path power selection threshold. If this occurs the path or paths are dropped.

Having chosen the appropriate paths, the receiver reconfigures and generates an output whose SNR (signal to noise ratio) is close to optimal. The receiver architecture now operates only on the selected branches. Assuming the delay spread is sufficient to require LMMSE operation, the adaptive coefficient update algorithm for the LMMSE filter aims to minimise the following function:

The final detection is made according to:

The filter weights are adaptively updated to achieve the minimum mean square error (MSE) during one symbol interval and this can be written as:

w (m) k,l + 2μe Hm)-(m)

(5)

Using this method it is possible to minimise the excess mean square error and hence significantly reduce the interference floor under all propagation conditions. Hence, the receiver is able to overcome prevailing interference and noise enhancement, even in deep fading channels. Since an adaptive algorithm is used and interference is suppressed before the multipath is combined, knowledge of the exact fading gain on each path is not required, however explicit information regarding multipath delay is necessary. The adaptive receiver does not require interference information and no training sequence is necessary, hence the method is considered suitable for use in both the UL and DL sections of the UMTS TDD standard. Furthermore, this reconfigurable adaptive LMMSE structure is attractive since it embodies both the fixed Rake and correlation architectures in a compact and simple receiver structure.

A full set of performance results can be found in [1][2]. Figure 5 presents a BER performance comparison between the proposed channel reconfigurable adaptive receiver and a number of fixed structures in multi-user time varying multipath channel.

Figure 5 demonstrates the dramatic performance enhancement offered using the Reconfigurable Adaptive Receiver (RAR) of the embodiment relative to the standard LMMSE receiver, rake receiver and chip level equaliser in a time varying multipath channel. The performance of the proposed reconfigurable receiver can be seen to be more stable than the LMS chip level equaliser and very close to that of the RLS chip-level equaliser. It should be noted that the RLS equaliser is considered too complex for practical implementation - generally being several orders of magnitude more complex than the newly proposed RAR. In the reconfigurable receiver, key channel parameters (μ_Ms, h_k, L) are obtained via channel estimation to determine the receiver's mode. The complexity of the proposed reconfigurable receiver is far less than that of a chip level equaliser and the technique can also be applied successfully to the UL.

Figure 6 demonstrates that the PS-LMMSE can significantly improve the performance and capacity of a multi-user TDD-CDMA system relative to the MF or Rake. At a BER of 2χl0^"3 and an Eb/N₀ of 22dB, the Rake and correlation receiver can support up to six users (37.5% of the processing gain). However, the PS-LMMSE can achieve up to nine users (56% of processing gain). For a system with three users at the lower Eb/N₀ value of 12 dB, the PS-LMMSE and NPS-LMMSE achieves a BER of 1.6χl0 \ which is worse than the Rake (BER of 10³). The MF offers the worst performance with a BER of 100^"2. These results show that both the PS-LMMSE and the NPS-LMMSE should be disabled for four or more users at Eb/N₀ levels less than or equal to 12dB.

The minimum required SNR grows as the number of users increases. However, the required minimum Eb/N₀ for the LMMSE is less than that of the MF for small numbers of users (four or less based on Figure 5). When more users enter the system, the value of Eb/(No+I₀) reduces, where I₀ represents the interference power spectral density. The PS and NPS-LMMSE are very sensitive to low values of SNR. This sensitivity arises from the iterative use of decision directed errors. The use of reconfigurability prevents the use of the LMMSE in channel with low S/N or low rms delay spread.

REFERENCES

The following list specifies the documents referred to in this document. Note that [1] and [2] are attached to the present document and form a part of the present document for the purpose of aiding the understanding of the invention but they should not be taken to impose limitations on the protection sought. [I] Y. Q. Bian and A. R. Nix, "Channel Adaptive Multiple Access and Intersymbol Interference cancellation techniques for TDD-Mode UMTS receivers," to be published at the IEE 3G Conference, London, March 26-28 2001

[2] Y. Q. Bian and A. R. Nix, "Performance Analysis of a Blind Adaptive

Linear Minimum Mean Squared Error Receiver for use in UMTS TDD-CDMA Basestations," to be published at the IEEE VTC conference, Rhodes, Greece, May 2001

[3] Proakis, 'Digital communications', 1989, International edition.

[4] Shimon, 'Multi-user detection for DS-CDMA communications', IEEE commun. magazine, October 1996, pp.124- 134.

[5] George L. Turin, 'The effects of multipath and fading on the performance of direct-sequence CDMA systems', IEEE JSAC, Vol. SAC-2, No. 4, July 1984, pp. 597-603.

[6] Dimitris, 'Data detection techniques for DS/CDMA mobile systems: A review', IEEE personal commun., June 2000, pp.25-34.

[7] Zvonar Z, 'Combined multiuser detection and diversity reception for wireless CDMA systems', IEEE Trans. Veh. Technol., 45, (1), 1996, pp. 205-211.

[8] Tero Ojanpera, 'Wideband CDMA for third generation mobile communications', 1998, Artech house.

[9] Matti, 'Advanced receivers for wideband CDMA systems', 1999, Oulu university library.

[10] 3G TS 25.102 vl.2.0, '3GPP UTRA (UE) TDD, Radio transmission and reception', July 1999.

[I I] R. Price, 'A communication technique for multipath channels', IRE Proc, March 1958, pp.555-570.

CHANNEL ADAPTIVE MULTIPLE ACCESS AND INTERSYMBOL INTERFERENCE CANCELLATION TECHNIQUES FOR TDD-MODE UMTS RECEIVERS

Y. Q. Bian and A. R. Nix

University of Bristol, Centre for Communications Research, Woodland Road, Bristol, BS8 1UB, UK

Abstract

This part of the document presents a comparison of simulation results for various Time Division Duplex (TDD) UMTS receiver architectures. Uplink and downlink simulations are performed over frequency-selective time varying channels. Results indicate that in practice, satisfactory performance cannot be obtained using fixed branch receiver architectures. To overcome this limitation, a reconfigurable blind adaptive Linear Minimum Mean Square Error (LMMSE) receiver is proposed to simultaneously track channel variations and mitigate the effects of Multiple Access Interference and Intersymbol Interference. This new method is compared with standard correlator, Rake receiver and chip level equalisation techniques. System performance is studied in terms of uncoded bit error rate for low and medium levels of rms delay spread. Simulations demonstrate that adaptive branch and parameter selection greatly reduces the interference floor in multi-user CDMA systems. More specifically, the proposed reconfigurable blind adaptive LMMSE is shown to offer competitive results at low relative complexity. This approach can be used to improve the performance and capacity of TDD mode UMTS networks.

1. INTRODUCTION

In Europe, the UMTS DS-CDMA (Direct Sequence Code Division Multiple Access) TDD mode will be used for licensed cellular operation in the 1900-1920 MHz and 2010-2015 MHz bands [12]. The system is primarily intended to offer 3G services in micro and pico cellular environments. In particular, the system intends to support traffic hot spots in campus style environments. The capacity of a TDD-CDMA system is divided asymmetrically between the uplink (UL) and downlink (DL). This asymmetry can be used to support multimedia services by adjusting the UL and DL requirements of the TDD-CDMA cell. The task of the receiver is to separate data belonging to the user of interest from data belonging to interfering users. In a TDD-CDMA system, ignoring the effects of multipath, DL interference is suppressed since the user codes are orthogonal and perfectly time aligned. However, in the UL the user codes are sent over independent, imperfectly time aligned, channels and significant cross correlation interference occurs. However, in the presence of multipath delay spread, UL and DL interference increases significantly. For this reason, independent UL (basestation receiver) and DL (mobile receiver) architectures and simulations are required.

In general, a Rake in CDMA is believed to represent the optimum direct sequence spread spectrum receiver for a single-user in a multipath environment [3]. However, for multi-user applications (as seen in practice), MAI (Multiple Access Interference) and the 'near-far problem' seriously degrade system performance [4]. Severe multipath fading is common in both dense urban and indoor environments. In a practical TDD- UMTS system, as mentioned previously, the reception of the desired user code suffers interference from the presence of other codes in the multipath propagation channel. The cross correlation noise of other users introduces a so-called interference floor, and the resulting system becomes interference limited, rather than noise-limited [5]. Unless suitable receiver processing is applied, the multipath channel will introduce MAI and thus limit the capacity and performance of a DS-CDMA system.

It is well known that Multi-User Detectors (MUD) offer significant improvements in DS-CDMA communications [4]. The optimal MUD takes the form of a Maximum Likelihood Sequence Estimator (MLSE). This structure offers the best performance and is resistant to the 'near- far' problem. However, the complexity of a MLSE receiver increases exponentially with the number of users present, making the implementation extremely difficult in practice. Therefore, over the last decade, most research in this area has focused on sub-optimal MUD solutions offering feasible implementation [6]. Examples include the multipath-decorrelating receiver and the LMMSE (Linear Minimum Mean Square Error) receiver. In the conventional multipath-decorrelating receiver, noise enhancement increases with the number of resolvable paths [7]. The adaptive LMMSE requires a symbol level training sequence to be sent from the desired user and this is a major limitation in the TDD-CDMA standard. Pre-Rake structures can be applied to either the UL or DL to support FDD (Frequency Division Duplex) and TDD operation. If operation is restricted to TDD mode then the complexity of pre-Rake terminals can be reduced [8]. A fixed (in terms of Rake branches) adaptive LMMSE- Rake is analysed in [9].

Due to natural time variations in both the mobile channel and the interference, satisfactory Bit Error Rate (BER) performance for a practical system cannot be obtained using a fixed-branch receiver structure. In order to improve system performance, a reconfigurable adaptive receiver for UMTS-TDD mode is proposed. In this part of the document, a number of different receiver architectures are studied and compared, including the correlation receiver, the Rake receiver and the blind adaptive LMMSE receiver. In addition, an adaptive DFE (Decision Feedback Equaliser) chip-level equaliser is analysed for DL use in the handset. Simulation results indicate that the best choice of receiver structure depends on channel characteristics such as the delay spread, path strength (relative to the strongest path) and signal to noise ratio for each resolvable multipath component. Given this situation, a low complexity reconfigurable channel adaptive receiver is recommended and evaluated in this part of the document.

2. SYSTEM DESCRIPTION

The simulation assumes the use of QPSK modulation. This modulation is sent over a frequency selective fading channel assuming K active users. An orthogonal Walsh code of length 16 is used in the spreading process. The wideband channel distortion is modelled using a time variant tapped delay line transformation. The impulse response of the UL multipath channel for the k-th user can be represented as: h_k(t) = ∑h_{I c}δ(t -τ_{l k} ) (1)

where h_k(t), L and x _t - represent the complex channel gain, the number of resolvable paths and the excess delay respectively.

The channel model consists of several path clusters (resulting from multiple reflection and diffraction). The fading processes on each path are statistically independent and Rayleigh distributed [10]. It is assumed that the channel multipath delay spread, T_m « L/W, is comparable or greater than the chip period T_c, where W represents the signal bandwidth. Hence, the multipath will introduce ISI (Inter Symbol Interference) and the propagation channel is affected by frequency selective fading. For a DS-CDMA system, this ISI will cause serious MAI. The received signal can be mathematically expressed as:

r( t) = ∑A_k ∑ ∑ h_ktb_kJs_k (t - iT„ -τ_kJ) + n(t) (2)

where 2M+1 is the number of data symbols and Ak, bk, Tb and S denote, respectively, the amplitude, data sequence, bit period and signalling waveform of the k-th user. n(t) represents the complex zero-mean background Additive White Gaussian Noise (AWGN) with power spectral density σ . At the receiver, the signal r(t) is passed through a matched Root-Raised Cosine (RRC) filter with a roll-off constant,©., of 0.22. The output from this filter is then sampled at the chip rate.

3. RECONFIGURABLE TDD RECEIVER

In this section a reconfigurable UMTS TDD mode detector is considered for the suppression of MAI and ISI over noisy, frequency selective, time varying channels. A novel blind adaptive LMMSE combined with dynamic parameter settings and path selection is proposed as a practical receiver offering high TDD-CDMA performance. The architecture of this reconfigurable adaptive receiver is shown in Figure 2 and its operation in Figure 4.

Firstly, mobile channel parameters such as rms delay spread and path power must be estimated to determine the value of the adaptive step-size, μLMs. If the rms delay spread value is low (e.g., below 75ns), the value of μu ts is set to zero. This implies that there is no need to support an adaptive LMMSE (mode 1) and the resulting receiver is reconfigured as a conventional Rake structure (mode 2). If only one significant multipath delay branch is detected, the architecture is reconfigured as a correlation receiver (mode 3). For larger values of rms delay spread, μuvis is set to an optimum value and blind adaptive LMMSE is performed. The adaption operates using decision directed (data-derived) training and aims to subtract MAI prior to multipath combining. MRC (Maximal Ratio Combining) and EGC (Equal Gain Combining) represent two common algorithms often used in this form of receiver. Compared with EGC, MRC offers a greater diversity gain and is widely employed in the Rake combiner [11].

Secondly, the strongest branch path, together with all other significant paths, are selected to determine the total number of branches (or taps) required for combining. The selected paths are continuously monitored to ensure that they do not fall below the relative path power selection threshold of -12dB. Having chosen the appropriate paths, the receiver reconfigures and generates an output whose SNR (signal to noise ratio) is close to optimal. The receiver architecture now operates only on the selected branches. Assuming the delay spread is sufficient to require LMMSE operation, the adaptive coefficient update algorithm for the LMMSE filter aims to minimise the following function:

ε^_kΛ\ -<,ή²} (3)

The final detection is made according to:

sigm±hl_j ) (4)

wiγ ⁾ = wff + Iμr™ (h_kJb? - yff )' = w["j> +

(5)

Using this method it is possible to minimise the excess mean square error and hence significantly reduce the interference floor under all propagation conditions. Hence, the receiver is able to overcome prevailing interference and noise enhancement, even in deep fading channels. Since an adaptive algorithm is used and interference is suppressed before the multipath is combined, knowledge of the exact fading gain on each path is not required, however explicit information regarding multipath delay is necessary. The adaptive receiver does not require interference information and no training sequence is necessary, hence the method is considered suitable for use in both the UL and DL sections of the UMTS TDD standard. Furthermore, this reconfigurable adaptive LMMSE structure is attractive since it embodies both the fixed Rake and correlation architectures in a relatively compact and simple receiver structure.

4. SIMULATION RESULTS

Uplink with medium delay spread:

Non reconfigurable receiver simulations were performed assuming K=8 users and a Power Delay Profile given by [-8.7506dB, -8.333 ldB, -30.3204dB]. The goal of a correlation receiver is to discriminate between the wanted and unwanted components in the receiver signal.

The correlation receiver is implemented as a bank of Matched Filters (MF) that correlate the received signal with the wanted user's code and maximises SNR by the resulting processing gain. However, this structure is only optimum in a non-fading single user environment.

Figure 7 demonstrates the UL simulation results for the full range of non-reconfigurable receiver structures. For the average path profile considered, the mean of the second path is marginally stronger than the first path, however both have similar relative strengths. MAI results from the asynchronous multi-user UL transmissions. This largely results from the independent nature of the multipath propagation channel, which enables the strongest path to occur at different times for different users. Figure 7(i) presents the convergence properties of the blind adaptive LMMSE. Taking just the strongest path, i.e. Path Selection (PS-1 tap), results in poor convergence and high residual MSE. The lowest MSE occurs when the two strongest paths are combined (PS-2 tap). For this channel, combining all three branches to cover the entire channel response, i.e No Path Selection (NPS), offers no benefit in terms of reducing the cross correlation noise floor or the MSE floor (in fact it increases slightly). These results confirm that adaptive path selection using a suitable power threshold is desirable. Figure 7(ii) shows the relative BER performance of the various UL receiver techniques. The single tap LMMSE offers the worst performance, as expected from the convergence results of Figure 7(i). The conventional correlation receiver offers the next level of performance, closely followed by the fixed (NPS) and adaptive PS Rake. Using Walsh coded, the correlation receiver will capture the signal from all multipaths and this adds to the cross-correlation noise to seriously degrade performance. Although the Rake outperforms the correlation receiver, the simulation results show an unacceptable irreducible BER floor of 10^"1. For the blind LMMSE, the selection of the single strongest path is clearly unacceptable. The selection of all three paths (NPS LMMSE) is significantly better than the Rake approach, however the dynamic selection of the two strongest paths produces the best result (LMMSE-2 tap). This result confirms that all paths within a suitable power window should be combined in the LMMSE.

Downlink with medium delay spread:

In the UMTS-TDD DL, a chip level training sequence is available to adapt the coefficients of a chip level equaliser. Figure 8(i) shows the convergence of a chip level DFE with 6 feedforward and 5 feedback taps. Two convergence algorithms have been simulated, the Least Mean Square (LMS) and the Recursive Least Square (RLS). As expected, the more complex RLS approach yields the fastest training and the lowest residual MSE. In addition, for some channel conditions, the LMS approach struggles to converge adequately within the 256-chip training period. Figure 8(ii) shows the various BER performance results for the DL analysis. Once again, the Rake outperforms the correlation receiver, however both the RLS chip level equaliser and LMMSE offer significantly superior performance. The graphs show that the blind adaptive LMMSE has a powerful ability to subtract more interference than a Rake for channels with significant rms delay spread. When suitable threshold parameters are employed to choose the correct number of LMMSE taps, the cross correlation noise floor can be removed (or significantly reduced), as shown in the case of the 2-tap LMMSE. Additionally, on the DL multiple users transmit synchronously through the same channel and individual power control is no longer required.

In the DL, MAI is essentially caused by the multipath channel. This implies that an adaptive chip-level equaliser can restore the orthogonality between user's codes. The chip-level equaliser requires timing information and a chip level training sequence. Fortunately, the existence of a 256 chip mid-amble code in the TDD-CDMA DL specification enables this technique to be applied. The blind adaptive LMMSE with dynamic path selection can achieve a BER of 10° at an E_t/N₀ of 19.5dB. This result is almost identical to the RLS driven chip level equaliser. The chip level equaliser with the LMS algorithm has an irreducible error floor of 2 10^"2. The path selection Rake exhibits an error floor of 3.5x10^" . For the case of medium rms delay spread values (e.g., greater than 75ns) and channels having approximately equal path strengths, the performance of the blind adaptive LMMSE is impressive. It should be noted that the adaptive blind LMMSE operates at the symbol level while the chip level equaliser operates at the far higher chip level.

Uplink with small delay spread:

Figure 9 shows the BER behaviour versus E_o/N₀ under low rms delay spreads conditions on the UL. The main signal energy arrives on the first path, with other weaker paths having much lower SNR.

For low values of rms delay spread, additional taps have no benefit in the Rake receiver. The results of Figure 9 indicate that the blind LMMSE is far more sensitive to low values of SNR than the correlation or Rake receivers (especially when the rms delay spread is low). This sensitivity arises from the use of blind or decision directed errors. As the value of E_b/N₀ increases, the blind LMMSE becomes feasible, offering excellent results at E_t,/N₀ values greater than 19 dB. In these types of channel, to obtain satisfactory performance at low Eb/N₀ levels, μL_Ms should be set to zero and weaker paths must not be selected (i.e. the receiver must dynamically reconfigure). Results show that at low delay spreads, the correlation detector has good overall UL performance.

Downlink with small delay spread:

The BER performance comparison for small DL rms delay spread conditions is given in Figure 10. Results demonstrate that simpler Rake combining using an adaptive path search threshold can achieve a better performance than more complex structures. This occurs since the coefficients do not need to adapt. The chip-level equaliser also achieves satisfactory performance, however these structures enhance noise at low EJN₀. The LMMSE does not perform well in such channels (requiring high E_b/N₀ to converge) and the receiver should be reconfigured in favour of Rake combining.

Reconfigurable Downlink with variable delay spread:

Figure 11 presents a BER performance comparison between the proposed channel reconfigurable adaptive receiver (see section 3) and a number of fixed structures.

The performance of the proposed reconfigurable receiver can be seen to be more stable than the LMS chip level equaliser and very close to that of the RLS chip-level equaliser. However, the RLS equaliser is considered too complex for practical implementation. In the reconfigurable receiver, key channel parameters (μLMS, hk, L) are obtained via channel estimation to determine the receiver's mode. The complexity of the proposed reconfigurable receiver is far less than that of a chip level equaliser and the technique can also be applied successfully to the UL.

5. CONCLUSIONS

In this part of the document several interference suppression and multipath combining schemes have been presented for UL and DL operation in the UMTS TDD mode. Simulation results demonstrate that the performance of the correlation detector and conventional Rake are often seriously corrupted by MAI and ISI. This distortion is strongly related to the characteristics of the mobile channel, with parameters such as rms delay spread and the signal to noise ratio for each resolvable multipath component playing a key role. Noise enhancement increases with the number of resolvable multipaths. Therefore, additional taps can increase the interference floor for a conventional correlation detector or Rake receiver. Moreover, the proposed blind adaptive LMMSE has been shown to be extreamly sensitive to the noise present on weaker paths. For channels with small rms delay spread or low SNR, results have shown that blind LMMSE decision directed training is not always desirable and poor convergence may occur. Hence, to overcome this limitation, a reconfigurable multimode receiver has been proposed, with adaptive mode and path selection based on channel estimation.

For both LMMSE and Rake operation, an adaptive receiver structure based on dynamic path searching has been used. In the case of the LMMSE, the receiver reduces both MAI and ISI over time varying, frequency selective, fading channels. The method is suitable for both UL and DL operation and complexity is far lower than alternative techniques. More specifically, for DL operation the reconfigurable LMMSE has far lower implementation complexity than the DFE chip-level equaliser. In conclusion, when suitable dynamic threshold parameters are used in the mode selection and choice of tap number and adaption constant, μi vis, impressive UL and DL performance can be obtained at low complexity with a flexible LMMSE based receiver structure. These properties make the method attractive for time varying multipath channels and the approach can be used to enhance the performance and capacity of TDD mode UMTS networks.

FURTHER REFERENCES

[12] 'The path towards UMTS technologies for the information society', No. 2 report from the UMTS Forum.

Performance Analysis of a Blind Adaptive Linear Minimum Mean Squared Error Receiver for Use in UMTS TDD-CDMA Basestations

Y. Q. Bian, A. R. Nix and J.P McGeehan

Centre for Communications Research, University of Bristol, Merchant Venturers Building, Woodland Road, Bristol, BS8 1UB, UK

Tel: +44 117 954 5203; Fax: +44 117 954 5206; email: Y.Q.Bian@bristol.ac.uk

Abstract

This part of the document presents receiver architectures and performance results relating to the UMTS (Universal mobile telecommunication system) TDD (time division duplex) mode. System performance in terms of uncoded BER (Bit Error Rate) and overall capacity are compared for MF (Matched Filter), Rake and blind adaptive LMMSE receivers. Uplink comparisons are performed in a multi-user scenario over time varying frequency selective channels. In order to achieve high performance, a modified blind adaptive LMMSE receiver is proposed. Factors such as channel variation, interference suppression and low implementation complexity are considered. The results demonstrate that the proposed receiver architecture can greatly reduce the interference floor at the basestation and thus significantly improve performance and capacity in the UMTS TDD mode.

1. INTRODUCTION

The TDD-CDMA (Time Division Duplex Code Division Multiple Access) system has been developed to offer third generation cellular services in the licensed 1900-1920 MHz and 2010-2015 MHz range [12]. The capacity of the system can be divided asymmetrically between the uplink (UL) and downlink (DL). TDD-CDMA is well suited to high rate multimedia type services. The peak data rate is at least 384 kb/s in microcells and 2 Mb/s in picocells (assuming unidirectional high rate transmissions) [13]

Multipath fading is common to both dense urban (microcell) and indoor (picocell) environments. The multipath channel degrades signal quality by introducing additional interference, thus limiting the capacity and performance of a DS (Direct Sequence) - CDMA system. The Rake receiver is believed to be the optimum DS-CDMA receiver for a single user in a multipath channel. However, in more realistic multi-user conditions, reception of the desired user code suffers interference from the presence of other user codes in the multipath channel. The cross correlation noise introduces an interference floor, and the resulting system becomes interference limited, rather than noise-limited [5]. For UL transmissions, the various mobile users are not perfectly time synchronous since each user communicates to the basestation (BS) through an independent, imperfectly time aligned channel. Each user becomes a source of interference to all other users. Therefore, practical receiver designs must consider the interference from other users in a multipath channel. Optimal multi-user detectors (MUDs) can free DS-CDMA systems from their well reported interference limitations [4], however they are generally too complex to implement. Over the last decade, research in this area has focused on the development of sub-optimal MUD solutions [4][6].

In a conventional multipath-decorrelating receiver [7], noise enhancement increases with the number of resolvable multipaths. The adaptive LMMSE [15] requires a predefined symbol level training sequence to be sent from the desired user to minimise the mean square error (MSE). Unfortunately, the TDD-CDMA standard does not support such a sequence. In addition, the LMMSE is known to suffer from phase error problems in channels suffering deep multipath fades [16]. A pre-Rake structure can be applied to either the UL or DL to enhance the performance of both FDD and TDD operation [8]. In order to improve system performance and overcome the limitations of the TDD- CDMA system, a blind adaptive LMMSE receiver with dynamic path selection (PS- LMMSE) is proposed and developed in this part of the document. The adaptation is based on time varying channel parameters such as signal to noise ratio (SNR) and RMS delay spread. Blind adaptation operates using decision directed (data-derived) training and aims to subtract Multiple Access Interference (MAI) prior to multipath combining. The evaluation of the BER performance and capacity versus receiver structure is presented here for a multi-user basestation.

2. SIGNAL MODEL

The simulation assumes the use of QPSK modulation. This modulation is sent over frequency selective fading channels assuming K simultaneously active users. Individual user data is aggregated at the BS. An orthogonal Walsh code of length 16 is used in the spreading process. The wideband channel distortion is modelled using a time variant tapped delay line transformation. The time-variant UL impulse response for the k-th user, h_k(f), can be written as:

_k (t) = I ∑ h_kJδ(t - τ_k ) 0)

where h_k,ι and T_k,ι represent the 1-th complex channel gain and excess delay respectively for the k-th multipath channel. L represents the maximum number of resolvable multipaths. The channel model consists of several path clusters (resulting from multiple reflection and diffraction). The fast fading processes on each path are statistically independent and Rayleigh distributed [10]. It is assumed that the channel multipath delay spread, T_m « L/W, is comparable or greater than the chip period T_c, where W represents the signal bandwidth. Hence, the multipath will introduce ISI (Inter Symbol Interference) and the propagation channel is affected by frequency selective fading. For the UL, the received waveform is generated as the summation of the signal from each user, which is calculated by the convolution of the user signal with its independent channel. The received signal can be mathematically expressed as: /. M r(t) = ∑ A ∑ Σ h_k,b_kJs_k(t - iT - τ_kJ) + n(t) (2) k=\ l=\>=-M

where 2M+1, A_k, h ,_\, T_s and S_k(t) denote, respectively, the number of data symbols, the transmit symbol amplitude for the k-th user, the i-th transmitted symbol in the k-th user data sequence, the symbol period and the signalling waveform of the k-th user. n(t) represents the complex zero-mean background Additive White Gaussian Noise (AWGN) with power spectral density σ . At the receiver, the signal r(t) is passed through a matched Root-Raised Cosine (RRC) filter with a roll-off constant, α, of 0.22. The output from this filter is then sampled at the chip rate.

3. PERFORMANCE ANALYSIS

The presence of a multipath channel substantially increases the need for MUD. The correlation receiver is implemented as a bank of Matched Filters (MF) that correlate the received signal with the wanted user's code and maximises SNR by the resulting processing gain. However, this structure is only optimum in a single user, non-fading environment. The conventional detector suffers from significant performance degradation in the presence of MAI, resulting in overall inefficiency in the communication system. For the case of simplicity, the following mathematical analysis is based on BPSK (although simulations make use of QPSK). The BER of a conventional receiver for the k-th user, p_c , can be approximated by [17]:

where N is the processing gain, Vk- defines the ratio of the received signal strength of the k'-th user to the received signal strength of the desired k-th user, and 2 ^β= f^" t - e^" ²/ dt (7)

In a single user environment, the Rake is a powerful technique for combating the effects of multipath fading. The error probability using a Rake at large SNR values (greater than 10 dB) can be approximated as [3]

where γ_c is the average SNR per channel and

'21 - 1 (2L - \)!

(9) Ll(L - l)!

Once multipath effects are introduced in a multi-user environment, due to the imperfect auto-correlation properties of the Walsh codes, multipath components become correlated and diversity gain is lost. Therefore, the cross correlation noise becomes channel-dependent. The BER of a multipath decorrelator for the k-th user is calculated as [18]:

where _{R (}i₎=—._Sk(t)s^ ^and H denotes the conjugate transpose function. In the absence of

N multipath, the k-th user is orthogonal to all other users and the decorrelator becomes equivalent to a single-user MF with R k⁼l. Otherwise, equation 10 clearly indicates that the decorrelating detector eliminates interference at the expense of noise enhancement, since R^(i)≥ \ . The MMSE detector has the crucial advantage of adaptive implementation. In this part of the document, the interference suppression scheme consists of replacing the matched filter by a Q-tap LMMSE, where the number of FIR (Finite Impulse Response) filters is equal to the processing gain (PG). Path selection is also necessary in this receiver structure. The output of each selected path takes the form:

Λ' (H)

where r^(,) represents the received signal vector over a processing window for the i-th symbol interval and y_k,ι^(l) represents the i-th symbol, on the 1-th branch, for the k-th user. W_k,ι are the complex coefficients of the Q-tap LMMSE detector for the 1-th branch of the k-th user. The coefficients are chosen to minimise the MSE, which depends on several random quantities, such as the cross-correlation, the time offsets, and the power levels of the received signals. The MSE is defined as:

Jk! = EψkA^bk - "A (12)

where i, is the hard BPSK decision for the k-th user, and is given by:

sigtf∑hljyϊj) (4)

In the simulation studies, the following QPSK hard decision rule is used for i_>

b['> h_kJy[' )] (13)

The filter weights are adaptively updated to achieve the minimum MSE during one symbol interval and this can be written as:

The BER of the adaptive LMMSE for the k-th user in a multipath channel can be calculated based on the Gaussian assumption [9] as shown below:

where σ is the variance of the desired response and Jk,ι is the final steady-state mean squared error for the adaptive algorithm. This equation can be used to estimate the impact of the noise and the interference floor on the receiver performance.

4. SIMULATION RESULTS

4.1 Medium RMS Delay Spreads (75-150ns)

Figure 12 demonstrates the UL simulation results for the different receiver structures in a medium RMS delay spread channel. K=8 users are assumed and the Power Delay Profile for the desired user is given by [-9.7dB, -10.7dB, -33.8dB]. The mean of the first path is marginally stronger than the second path, however both have similar relative strengths. MAI will result from the asynchronous multi-user UL transmissions. This largely results from the independent nature of the multipath propagation channels, which enables the strongest path to occur at different times for different users.

In Figure 12, the relative BER performance of the various UL receiver techniques is presented. The single tap LMMSE offers the worst performance since insufficient signal energy can be captured from the first path alone. The conventional correlation receiver offers the next level of performance, closely followed by the fixed (NPS) and adaptive Rake with path selection (PS). As discussed in the previous section, the use of Walsh spreading codes means the correlation receiver will capture the signal from all multipaths and thus add to the cross-correlation noise, which seriously degrades performance. The Rake outperforms the correlation receiver (however diversity gain is lost due to partially correlated multipaths), achieving a BER of 3χl0^"3 at 22dB E_b/N₀. For the blind LMMSE, the selection of the single strongest path is clearly unacceptable. The selection of all three paths (NPS-LMMSE) is significantly better than the Rake approach for E_b/N₀ values greater than 23 dB. The dynamic selection of the two strongest paths produces the best result (PS-LMMSE-2 tap). This result confirms that all paths within a suitable power window (12 dB in this case) should be combined in the LMMSE. To achieve a target BER of 10^"2, the required E_b/N₀ for the PS-LMMSE (around 21.5dB) is less than that for the NPS-LMMSE (around 22.5dB). As the E_b/N₀ increases, the NPS-LMMSE has a comparable performance to that of the successful PS- LMMSE. These results indicate that adaptive path selection using a suitable power threshold is desirable.

The convergence properties of the blind adaptive LMMSE at an E_b/N₀ of 22dB are shown in Figure 13. Taking just the strongest path, i.e. Path Selection (1 tap), results in poor convergence and high residual MSE. The lowest MSE occurs when the two strongest paths are combined (2 tap). For this channel, combining all three branches will cover the entire channel response. However, NPS offers no benefit in terms of reducing the cross correlation noise floor or the MSE floor. From inspection of Figure 12, the use of NPS causes unreliable decisions at E_b/N₀ values less than or equal to 22dB. This is to be expected from the convergence results shown in Figure 13. 4.2 BER Performance versus Number of Users

Figure 14 demonstrates that the PS-LMMSE can significantly improve the performance and capacity of a multi-user TDD-CDMA system relative to the MF or Rake. At a BER of 2x 10^'3 and an E_b/N₀ of 22dB, the Rake and correlation receiver can support up to six users (37.5% of the processing gain). However, the PS-LMMSE can achieve up to nine users (56% of processing gain). For a system with three users at the lower E_b/N₀ value of 12 dB, the PS-LMMSE and NPS-LMMSE achieves a BER of 1.6xl0^"3, which is worse than the Rake (BER of 10^"3). The MF offers the worst performance with a BER of 10^'2. These results show that both the PS-LMMSE and the NPS-LMMSE should be disabled for four or more users at Eb/N₀ levels less than or equal to 12dB.

The minimum required SNR grows as the number of users increases, as shown in Figure 15. However, the required minimum Eb/N₀ for the LMMSE is less than that of the MF for small numbers of users (four or less based on Figure 15). When more users enter the system, the value of Eb/(No+I₀) reduces, where I₀ represents the interference power spectral density. The PS and NPS-LMMSE are very sensitive to low values of SNR. This sensitivity arises from the iterative use of decision directed errors.

4.3 Low RMS Delay Spreads (less than 75ns)

Figure 16 shows the UL BER behaviour versus Eb/N₀ for low rms delay spread conditions. The main signal energy arrives on the first path, with other weaker paths having much lower SNR. For low values of rms delay spread, additional taps have no benefit in the Rake receiver. The results of Figure 16 indicate, once again, that blind LMMSE is far more sensitive to low values of SNR than either the correlation or Rake receiver. However, as the value of E_b/N₀ increases, the blind LMMSE becomes feasible, offering excellent results for E_b/N₀ values of 19.5 dB or higher. This implies that a reconfigurable receiver is required, using the LMMSE at high E_b/ ₀ values and the MF or Rake at lower values [1]. Results show that at low delay spreads, the required minimum E_b/N₀ at a target BER of 10^" (15.5dB for the correlation receiver) is much lower than for larger rms delay spread channels (19-20 dB from Figure 12). As mentioned above, in these channels the correlation detector offers good overall UL performance.

5. CONCLUSIONS

In this part of the document several interference suppression and multipath combining schemes have been presented for UL operation in the UMTS TDD mode. Simulation results demonstrate that the performance of the correlation detector and conventional Rake are often seriously corrupted by MAI and ISI. This distortion is strongly related to the characteristics of the mobile channel, with parameters such as rms delay spread and the signal to noise ratio for each resolvable multipath component playing a key role. Noise enhancement increases with the number of resolvable multipaths. Therefore, additional taps can increase the interference floor for the Rake structure and multipath- decorrelator. Moreover, the proposed blind adaptive LMMSE has been shown to be extremely sensitive to the noise present on weaker paths. For channels with small rms delay spreads or low SNRs, results have shown that blind LMMSE decision directed training is not always desirable since poor convergence may occur. An adaptive receiver structure based on dynamic path searching has been used in this part of the document to improve performance. The blind adaptive LMMSE receiver reduces both MAI and ISI over frequency selective time varying channels. The method is suitable for use on both the UL and DL of TDD-CDMA. Further performance gains can be obtained by reconfiguring the LMMSE as a Rake or correlation receiver at low SNR and/or RMS delay spreads [1].

In conclusion, when suitable dynamic threshold parameters are used in the choice of tap number, impressive UL performance can be obtained with a flexible LMMSE based receiver structure. These properties make the method attractive for time varying multipath channels and the approach can be used to enhance the capacity and performance of a TDD mode UMTS network.

FURTHER REFERENCES

[13] Prasad, Ramjee, 'Third generation mobile communication systems', 2000, Artech house.

[14] Verdu, S., 'Minimum probability of error for asynchronous Gaussian multiple- access channels', IEEE Trans. Inform. Theory, Vol. IT-32, pp.85-96, Jan.1986.

[15] U. Madhow and M. L. Honig, 'MMSE interference suppression for direct- sequence spread-spectrum CDMA', IEEE Trans. Comm., Vol. 42, pp3178-3188, No. 12, Dec. 1994.

[16] Afonso N. Barbosa, 'Adaptive detection of DS/CDMA signals in fading channels', IEEE Trans. Comm., Vol. 46, ppl 15-124, No.l . Jan. 1998. [17] Scott L. Miller, ' An adaptive direct-sequence code-division multiple-access receiver for multiuser interference rejection', IEEE Trans. Comm. Vol.43, No 2/3/4, 1995, ppl746-1754

[18] Sergio Verdu, 'Multiuser detection', 1998, Cambridge University Press.

Claims

1. A receiver for receiving CDMA-type signals, the apparatus being arranged to despread and demodulate signals received on a channel by using one of a plurality of reception processes, the receiver being arranged to select one of said reception processes for use with said received signals on the basis of a measured property of the channel.

2. A receiver according to claim 1, wherein said receiver comprises a number of rake elements, the signals on which are combined in the process of signal demodulation and despreading.

3. A receiver according to claim 2, wherein the number of said rake elements used in the reception process is set in dependence upon the delay spread of the received channel.

4. A receiver according to claim 2 or 3, wherein each rake element includes an adaptive filter.

5. A receiver according to claim 4, wherein the adaptation rate of each filter is altered in accordance with the measured property of the channel.

6. A receiver according to claim 5, wherein adaption rate adjustment is made on the basis of a signal to noise-plus-interference ratio for the channel.

7. A receiver according any preceding claim, wherein said reception process is selected on the basis of the delay spread within said channel.

8. A receiver according to any preceding claim, wherein said reception process is selected on the basis of a signal to noise-plus-interference ratio for the channel.

9. A receiver according to any preceding claim, wherein said receiver is further arranged to select one of said reception processes for a given user by performing a channel estimation process on the channel being received for that user.

10. A receiver according to any preceding claim, wherein said channel property comprises one or more of rms delay spread and channel path power.

11. A receiver according to any preceding claim, wherein said reception processes include a correlation mode, a rake mode and a LMMSE mode.

12. A receiver according to any one of claims 4 to 6, wherein at least one of the adaptive filters performs a LMMSE process.

13. A method of receiving CDMA-type signals, the method comprising selecting one of a plurality of reception modes to demodulate and despread signals received on a channel in accordance with a property of said channel.

14. A method according to claim 13, wherein a number of components of said received signals, which differ from one another by relative path delays, are selected for use in demodulating and despreading said received signals.

15. A method according to claim 14, wherein, said components are selected by comparing the received signal to a threshold.

16. A method according to claim 15, wherein where several of said components exceed the threshold, the number of components used is limited.

17. A method according to claim 16, wherein the number of components is limited by selecting a predetermined number of the strongest components.

18. A method according to any one of claims 14 to 17, wherein where only one component is selected the selected demodulation and despreading mode comprises a correlation mode.

19. A method according to any one of claims 14 to 18, wherein if a signal to noise-plus-interference ratio is greater than a given threshold and a delay spread is greater than another threshold then the selected demodulation and despreading mode comprises an LMMSE mode.

20. A receiver for receiving CDMA-type signals substantially as described herein with reference to, or as shown in, Figures 2 to 6.

21. A method of receiving CDMA-type signals substantially as described herein.