T4B.2
A Self-reconfigurable Receiver Architecture for Software Radio Systems Henrique C. Miranda, Pedro C. Pinto, Sérgio B. Silva FEUP / INESC Porto, Rua Dr. Roberto Frias, s/n 4200-465 Porto (Portugal) Phone: +351 22 5081474, Fax: +351 22 2094250, E-mail:
[email protected]
Abstract – This paper proposes an adaptive digital receiver architecture suitable for Software Defined Radio systems. This receiver is prepared to handle modulation schemes whose symbols belong to a linear, twodimensional signal space (such as M-PSK or M-QAM). It is capable of identifying key signal parameters (e.g., symbol rate and constellation) and, in an autonomous way, modify its operation accordingly. Four aspects of selfreconfiguration are addressed: baseband filtering, symbol synchronization, carrier synchronization and constellation recognition. The performance of the receiver is evaluated based on simulation results, and several usage scenarios for the proposed architecture are envisaged. Keywords – software radio, self-reconfiguration, synchronization, constellation recognition, fuzzy clustering
1 Introduction Several authors [1][2] have tried to develop intelligent receivers that are capable of identifying the incoming modulation scheme and reconfigure themselves accordingly. However, these architectures support only a very limited number of constellations, lacking both generality and performance. In this paper, an adaptive digital receiver architecture which is suitable for Software Defined Radio systems is presented. This receiver is prepared to handle not only M-PSK and M-QAM schemes, but also other modulation schemes whose symbols belong to a linear, two-dimensional signal space (e.g., nonuniform or fractal constellations). Thus, the proposed architecture achieves high generality and acceptable performance, being able to identify key signal parameters such as symbol rate and constellation, and self-reconfigure accordingly. This architecture can be used in receiver applications where spectrum surveillance and management can be carried out automatically and independently of signal properties. Moreover, in electronic warfare scenarios, electronic counter measures such as jamming can be applied more efficiently if the target signal characteristics are recognized. In civilian applications such as mobile communications, since the modulation properties of the transmitted signal can be identified, mobile terminals using different air interfaces can be served by a single multi-standard adaptive base station. This enables roaming of clients through geographical zones with different adopted standards, without modification of the mobile terminal. Section 2 presents the proposed receiver architecture, focusing on the four main aspects of self-reconfiguration: baseband filtering, symbol synchronization, carrier synchronization and constellation recognition. The receiver filter response adaptation is based on parameters fed by a symbol rate estimator block. The symbol synchronizer is responsible for sampling the baseband
signal at the optimal instants, while the carrier synchronizer prevents the constellation from rotating due to a carrier frequency error. The symbol decision block is aided by a constellation identifier, which estimates the optimal decision regions for the received signal modulation. All these blocks are able to operate independently of the modulation scheme. Section 3 presents simulation results for several blocks of the proposed receiver, in order to evaluate their performance, advantages and limitations. Section 4 exemplifies where the proposed receiver can be located in the usual communications protocol stack, and how it can interact with the other protocol layers. Since this receiver layout is very demanding in terms of computational power, it was implemented on a cluster of computers, interconnected by a high-speed network. The signal processing software was developed under the GNU Radio 1 framework. The LAM2 implementation of the MPI standard was used to provide transparent internode communication.
2
Proposed receiver architecture
The proposed receiver architecture is illustrated in the diagram of Fig. 1, revealing the information flow among its blocks. On a multi-mode adaptive receiver, the use of a dynamic filter is of utmost importance, as different received signals often have different spectral characteristics. As shown in Fig. 1, the estiˆ is used to dynamically adjust the baseband mated symbol rate, R, filter bandwidth. The symbol rate estimator operates by applying a nonlinearity (magnitude-squared) to the complex baseband signal, which causes a spectral line to appear at the symbol rate. A first estimate of R is obtained by calculating a moderate size FFT (16k points) of the resulting signal and then locating the frequency of the spectral peak. This estimate can be refined by applying a CZT (Chirp-Z Transform) to the region near the spectral peak. This way, the symbol rate estimation error can be greatly reduced. The purpose of the symbol synchronizer (Fig. 1) is to determine the optimal symbol sampling instants for the incoming signal. It accepts a complex baseband signal with several samples per symbol and produces a stream of correctly sampled complex symbols (one sample/symbol). Several well-known symbol synchronizers can be used for this purpose, as long as they are inde1 http://www.gnuradio.org 2 http://www.lam-mpi.org
complex−signal flow real−signal flow configuration−signal flow
constellation identifier phase rotator
IF signal
configurable FIR filter
symbol synchronizer
signal quality, modulation scheme
constellation centroids symbol detector
estimated symbols
FIR coefficients NCO
symbol rate estimator
^ R
coefficients generator
carrier synchronizer
Figure 1: Receiver block diagram. pendent of the modulation scheme. Three different synchronizers were implemented and have proved to perform well. The first is a spectral line synchronizer, whose main advantage is simplicity, although it requires several samples/symbol (10 or more) to achieve an acceptable level of timing jitter. Another alternative is a symbol synchronizer operating in a feedforward configuration, based on the combined use of group delay estimation and interpolation. As an example, a synchronizer using the Lee estimator [3] is able to work with only two samples/symbol but needs additional logic to handle symbol skipping/stuffing operations, due to the incommensurability of the T symbol /Tsampling ratio. Finally, it is also possible to use a feedback structure combined with interpolation [4]. The purpose of the carrier synchronizer (Fig. 1) is to provide a closed-loop control system that compensates for frequency mismatches (∆f ) between the received carrier signal and the local NCO (Numerically Controlled Oscillator). This ensures that the resulting signal constellation does not rotate. The carrier synchronizer follows a decision-directed approach using the wellknown maximum-likelihood phase synchronizer described in [5], which is governed by the following set of equations: ˆ + 1) = θ(k) ˆ + γe(k) θ(k , (1) ˆ e(k) = Im cˆ∗ x(k)e−j θ(k) ˆ where θ(k) is the estimated phase, γ is the iteration step and cˆ is the reference symbol nearest to x(k), the input symbol. However, the γ and cˆ parameters are highly dependent on the modulation scheme. To solve this dependence, the previous equations were modified to: ˆ + 1) = θ(k) ˆ + γe(k) θ(k , (2) ˆ e(k) = arg c∗ x(k)e−j θ(k) where γ is now a fixed parameter and c = ±1 ± j. The use of arg {.} instead of Im {.} ensures that γ can be chosen in way that makes synchronization independent of the input signal amplitude and modulation scheme. The use of c = ±1 ± j means that initially, the symbol detector is optimally configured for QPSK, using the real and imaginary axis as the decision boundaries for the symbol estimates. However, the carrier phase synchronizer can still prevent phase rotation even if the constellation is not QPSK (e.g., BPSK, 8-PSK and 16-QAM), but at the expense of a higher phase error variance and higher frequency offset sen-
sitivity. After the constellation identifier produces the correct decision regions, these will be used by the symbol detector to determine which of the M reference centroids is closer to the incoming symbol (in terms of signal space Euclidean distances), thus producing a maximum-likelihood symbol estimate. At this stage the tracking performance is best for the incoming constellation. The constellation identifier is used to estimate several parameters that characterize the received two-dimensional constellation. These parameters are: the number of symbols of the constellation, M; the coordinates of these M symbols and the S/N of the received signal. The first two parameters are then used to determine the modulation scheme (e.g., M-PSK or M-QAM). The identifier accepts at the input a block of N clust complex symbols, belonging to a non-rotating constellation (ensured by the carrier synchronizer) and sampled with the correct timing (ensured by the symbol synchronizer). It then applies a fuzzy c-means (FCM) clustering algorithm to these symbols, with the number ofcentroids to be detected set to C = 2, 4, 8, . . . , 2M , . . . , 2K . The parameter K is determined by the maximum expected number of centroids of the received constellation. Notice that a large K also requires a large number of input symbols, N clust , so that the constellation centers are accurately detected. Subsequently, a minimum hard-tendencies validity indicator [6] is calculated for each of these K sets of detected centroids, and the set that maximizes this indicator is chosen. This yields the constellation size M, as well as the corresponding symbol coordinates. As it is well known, FCM algorithms often have problems in correctly determining the cluster centers, because they can converge to a final (incorrect) set of centroids that locally maximizes the objective function (not globally, as desired). As a result, these algorithms are very sensitive to initial center choices, requiring proper initialization to ensure that correct results are obtained. To solve this problem, the min-max-median algorithm proposed in [7] was successfully employed. This allows the FCM algorithm to be initialized with well-separated centers that are already close to the final centers, thereby ensuring proper convergence and reducing the number of necessary iterations. After determining the coordinates of the constellation centers, Ci ∈ C, i = 1 . . . M , an estimate of the signal quality (S/N ) can be obtained. In essence, the signal power is simply easily S = |Ci |2 , i = 1 . . . M , where denotesthe average operator, and the noise power is N = |Ik − Ck |2 , k = 1 . . . Nclust ,
where Ik is a symbol taken from the block of N clust input symbols and Ck the corresponding (i.e., nearest) constellation symbol. Moreover, note that symbol timing errors and carrier synchronizer phase jitter are also included in this calculation of the noise power. Once coordinates of the constellation centers, C i , are known, several statistics such as the number and location of the C i are calculated and compared to a table of properties for all the expected modulations, thus identifying the current modulation scheme. Furthermore, the communications standard can also be identified by looking up a table that associates the several standards with the corresponding modulation schemes and symbol rates. The constellation identifier may operate according to the state diagram of Fig. 2, in which the receiver periodically identifies the input constellation and loads the corresponding protocol stack when a new scheme is detected L consecutive times.
Table 1: BPSK error performance for several transmitter filter roll-off factors. α 0.1 0.3 0.7 0.5 (optimum)
Es /N0 (dB) 5 7 0.0065 0.0012 0.0061 0.0009 0.0061 0.0008 0.0060 0.0008
Table 2: Maximum supported frequency error (∆f /R), as a function of the modulation scheme and E s /N0 (γ = 0.1). Modulation BPSK QPSK 8-PSK 16-QAM
Unknown scheme, analyze constellation with QPSK sync Same scheme detected L consecutive times Load protocol stack
3 0.025 0.023 0.022 0.022
10 0.4 % 0.4 % – –
Es /N0 (dB) 15 20 25 0.7 % 0.9 % 1.0 % 0.7 % 0.9 % 1.0 % 0.2 % 0.3 % 0.3 %
30 1.0 % 1.0 % 0.3 %
0.1 %
0.2 %
0.2 %
0.2 %
Different schemes detected L consecutive times
True Known scheme, analyze constellation with optimized sync
Figure 2: State diagram of the constellation identifier.
3 Simulation results The following subsections present the results of several simulations for the adaptive filter, carrier synchronizer and constellation identifier, using the M ATLAB /S IMULINK environment.
3.1 Adaptive FIR filter performance To evaluate the performance of the adaptive filter, several simulations were performed using a square-root raised-cosine filter with various roll-off factors at the transmitter side. The receiver employs the same type of filter with a fixed roll-off factor of α = 0.5 and a dynamically adjusted bandwidth. As shown in Table 1, the degradation of the system performance for BPSK is not significant when maintaining a fixed receiver baseband filter roll-off and varying only its bandwidth. The loss in performance is only noticeable with lower values of transmitter filter roll-offs. Similar results can be expected for denser modulation formats.
3.2 Adaptive carrier synchronizer performance To evaluate the carrier synchronizer robustness to frequency errors and noise, the maximum allowable relative frequency error (defined as ∆f /R) was measured for several modulation
schemes and Es /N0 ratios, as shown in Table 2. As expected, the carrier synchronizer works best with BPSK and QPSK, exhibiting maximum allowable frequency errors up to 1% of the symbol rate. Additionally, note that the parameter γ is set to a fixed optimized value, which ensures synchronization for all the intended modulation schemes.
3.3
Constellation identifier performance
Fig. 3 (a) depicts the performance of the constellation recognizer, in terms of percentage of correct identifications. Two situations are compared: first, the carrier synchronizer is disabled and the frequency error is zero; second, the carrier synchronizer is enabled and the frequency error is 0.1%. Clearly, the carrier synchronizer introduces a performance loss of about 2 dB for 16-QAM, in terms of constellation recognition. This occurs because it has been assumed that the system is still in an initial state, with the carrier block configured with QPSK decision boundaries. Consequently, the carrier synchronizer introduces a phase jitter in the constellation, as shown in Fig. 3 (b), which compromises the recognition process with higher order schemes (e.g., 256-QAM).
4
Usage scenarios
The proposed digital receiver can be used in two distinct protocol-stack configurations, as shown in Fig. 4. This paper has so far assumed the usage scenario of Fig. 4 (a), where the proposed receiver is used at Layer 1 as a universal demodulator (i.e., independent of the communications standard). The Symbol Demapper Layer accepts the M-ary symbols from the Demodulator Layer and converts them into bits. Both the Symbol Demapper and Data Link layers are highly dependent of the communications standard, so they must be dynamically loaded according
Communications standard 1
Communications standard 2
Data link 1
Data link 2
Symbol demapper 1
Symbol demapper 2
100 without phase jitter with phase jitter
L2
95
bits QPSK
90
16−QAM
L1
M−ary symbols
% of correct identifications
85
Universal demodulator 80
75
(a) Proposed demodulator common to the upper layers of all protocol stacks.
70
65
Communications standard 1
60
Communications standard 2
55
L2 50 −4
−2
0
2
4 6 E /N (dB) s
8
10
12
Data link 1
Data link 2
Symbol demapper 1
Symbol demapper 2
Optimized demodulator 1
Optimized demodulator 2
bits
14
0
L1 Comm. standard identifier
M−ary symbols
(a) Percentage of correct identifications (Nclust = 5000 and L = 1). Protocol stack loading
1.5
(b) Proposed demodulator used only for communications standard identification.
1 0.5 0
Figure 4: Proposed configurations for the receiver protocol stack.
−0.5
Fig. 4 (b) is used. Future developments on the proposed receiver architecture −1.5 would include the refinement of some receiver blocks (espe−1.5 −1 −0.5 0 0.5 1 1.5 cially the carrier synchronizer) in order to support higher-order modulation schemes, such as 256-QAM. Also, further work (b) Phase jitter introduced by the carrier synchronizer (Es /N0 = 30 dB, is needed to include other commonly used digital modulation γ = 0.1, ∆f /R = 0.1 %). schemes (e.g., CPM modulations) in order to create a fully adaptive, flexible, software radio receiver. −1
Figure 3: Simulation results for the constellation identifier.
References to the information provided by the constellation identifier (Universal Demodulator Layer). In Fig. 4 (b), the proposed demodulator is used only for the identification of the communications standard, after which a complete, dedicated protocol stack is loaded. This configuration has several advantage over the one in (a), namely in terms of flexibility and performance, since a highly optimized demodulator can be used to produce the M-ary symbols to the upper layers, instead of the generic and less optimized demodulator in (a).
[1] K. Nolan, L. Doyle, D. Mahony, P. Mackenzie, Signal Space based Adaptive Modulation for Software Radio, NTRG, Trinity College.
[2] B. Mobasseri, Digital Modulation Classification using Constellation Shape, ECE Department, Villanova University, 1997.
[3] S. Lee, A New Non-Data-Aided Feedforward Symbol Timing Estimator Using Two Samples per Symbol, IEEE Communications Letters, vol. 6, pp. 205-206, May 2002.
[4] F. Gardner, Interpolation in Digital Modems – Part I: Fundamentals, IEEE Transactions on Communications, vol. 41, no. 3, March 1993.
[5] U. Mengali, A. Andrea, Synchronization Techniques for Digital Receivers,
5 Conclusions and future developments The main advantages of the proposed receiver architecture are generality (within the two-dimensional signal space) and adaptability. However, such generality is obtained at the cost of performance, but this aspect is not important if the configuration of
Plenum Press, 1997, chapter 5.
[6] F. Rivera, E. Zapata, J. Carazo, Cluster Validity based on the Hard Tendency of the Fuzzy Classification, Pattern Recognition Letters, 11:7-12, 1990.
[7] Xue-wen Chen, Clustering Gene Expression Data with Min-Max-Median Initialized Fuzzy C-Means Algorithms, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign.
