PERFORMANCE OF NEW ATTACHED SYNC MARKERS FOR TURBO-CODE FRAME SYNCHRONIZATION IN DEEP-SPACE TELEMETRY SCENARIO Ricard Abellò 1, Paolo Andreazza 2, Noureddine Boujnah 2, Gian Paolo Calzolari 1, Xavier Enrich 3, Roberto Garello 2 1 European Space Agency, ESOC, Robert Bosch Str.5, 64293 Darmstadt, Germany Email: ricard.abello@esa.int, gian.paolo.calzolari@esa.int. 2 Politecnico di Torino, Dipartimento di Elettronica, Corso Duca degli Abruzzi, 24, 10129 Torino Italy Email: paolo.andreazza.84@gmail.com, garello@polito.it, noureddine.boujnah@polito.it. 3 Callisto Ltd., European Space Agency, ESOC,Robert Bosch Str.5, 64293 Darmstadt, Germany Email: xavier.enrich@esa.int I. INTRODUCTION Space systems and technologies for deep-space communication are very demanding activities for resource and research. They are regulated by the recommendations of Consultative Committee for Space Data Systems (CCSDS), which allows the standardization and the diffusion of state of art technology in this field. Turbo codes are currently the most powerful CCSDS coding option, with performance very close to Shannon s channel capacity theorem. The synchronization of turbo-encoded data between transmitter and receiver is achieved by using a stream of fixed-length Codeblocks (or Transfer Frames) with an Attached Synchronization Marker (ASM) between them. These sequences,, are appended to telemetry transfer frames in the transmitter side in order to be detected and monitored by the frame synchronizer at the receiver side, thus identifying the correct position of the Turbo Codeblocks within the received bit stream. In this work we evaluate new sequences with synchronization properties better than the ones currently shown by CCSDS ASMs, for each of the four code rates... When considering a practical scenario, including noise and quantization, the frame synchronizer must work by setting a proper threshold. This threshold will determine the of False Lock and Missed Detection probabilities. By simulation, the threshold has been optimized for operating points corresponding to a typical of Frame Error Rate (FER) equal to 1e-4. The Receiver Operating Characteristics (ROC) performance of the studied synchronization sequences show that strong improvements can be achieved with respect to the current CCSDS ASMs. II. CCSDS RECOMMENDATIONS Space-to-ground Telemetry links are regulated by CCSDS recommendations, adopted by the most relevant Space Agencies all around the world. Though different applications (deep space and near Earth research, Earth observation, meteorological, tracking and data relay missions) have different requirements, CCSDS recommendations aim to establish a general framework in terms of communication links, protocols and data types. The current CCSDS Telemetry Synchronization and Channel Coding Recommendation [1] include the following coding options: Reed Solomon codes. Convolutional Codes. Serial Concatenation of a Reed-Solomon and a convolutional code. Turbo codes. Currently, turbo codes are the most performing option included into the standard: they are the best solution for deep space missions where link-budget constraints are severe. CCSDS turbo codes allows for four different codeblock lengths (k = 1784, 3568, 7136, and 8920 bits) that can be combined with four different code-rates (r = 1/2, 1/3, 1/4, and 1/6). More details on their performance can be found in [2] and [3]. CCSDS turbo codes have already been applied in space. For ESA, SMART-1 was the first mission using turbo codes. It aimed at testing new technologies for space communications, including turbo codes, and making investigations on moon nature. The spacecraft was launched in September 2003 and entered the moon orbit in November 2004. In September 2006 the mission ended and the spacecraft impacted on the moon. The onboard Ka-band Telemetry Experiment (KaTE) instrument made experimental turbo coded transmissions using rate 1/4; encoded data were received by ESA's 12 meters antenna in Villafranca de Castillo (Spain) and successfully decoded. Another example of application will be Bepi Colombo, a joint mission between ESA and the Japanese Aerospace Exploration Agency (JAXA). Two different spacecrafts are going to be launched in 2014 and they should reach
Mercury in the middle of 2020. ESA will build the Mercury Planetary Orbiter whereas JAXA the Mercury Magnetospheric orbiter. The ESA orbiter will study the surface and the inner composition of Mercury, the latter its surrounding Magnetosphere. This mission will shed light on Mercury's nature and the solar system evolution. The entire CCSDS turbo code family will be available for a reliable transmission from such a remote location. ESA's 35 m antenna in Cebreros (Spain), together with other Japanese facilities, will be used as ground station. III. CODEWORD SYNCHRONIZATION BY ASM At the receiver side, before starting the actual decoding process, the received stream shall be delimeted such that the turbo decoder exactly locates the symbols corresponding to each turbo codeword. CCSDS turbo codeblock synchronization is achieved through Attached Sync Markers (ASM) sequences. The frame synchronizer is the device in charge of identifying the ASM pattern during the acquisition phase and to check its presence at regular intervals. CCSDS standard [1] provides ASM patterns for all the possible turbo-code rates. In particular, the ASM length N is fixed equal to 32/r, where r is the code-rate. In the following Tab. 1 we report the ASM patterns recommended into the standard. Tab. 1. The CCSDS Attached Sync Markers Code-rate r ASM length N [bits] ASM (hex code) 1/2 64 034776C7 272895B0 1/3 96 25D5C0CE 8990F6C9 461BF79C 1/4 128 034776C7 272895B0 FCB88938 D8D76A4F 1/6 192 25D5C0CE 8990F6C9 461BF79C DA2A3F31 766F0936 B9E40863 IV. THE APERIODIC AUTOCORRELATION FUNCTION OF CCSDS ASM The most important figure of merit of an ASM is the (aperiodic) autocorrelation function, defined as: N R bb i i i 1 where b i is the bipolar representation of each single bit forming the ASM, and is the offset. The ASM performance depends on the magnitude of side lobes with respect to the correlation peak (which is equal to N for τ=0). We then define the Peak-Side Lobe (PSL) parameter as max R 0 PSL 20log N It is important to note that, when 2-PSK modulation is employed, codeword synchronization is acquired before solving phase ambiguity. This is the reason why negative peaks must be taken into account, too, and the numerator is computed on the absolute of R(τ). A. THE CCSDS ASMs WITH N=64 AND 96 BITS The CCSDS ASM patterns of length 64 and 96 bits were found using neural network search algorithms [4]. The autocorrelation functions for 64 and 96 bits CCSDS ASMs are shown in Fig.1 and Fig.2, and their secondary peak s are listed in Tab.2 and Tab.3, respectively.
Fig. 1. The autocorrelation function of CCSDS ASM(64) Tab. 2. of secondary peaks and s for CCSDS ASM(64) ±8 +6 ±48, ±32, ±28, ±26, ±18, ±6, ±2-6 Fig. 2. The autocorrelation function of CCSDS ASM(96) Tab. 3. of secondary peaks and s for CCSDS ASM(96) ±52, ±40, ±6 +8 ±14, ±2-10 B. THE CCSDS ASMs WITH N=128 AND 192 BITS The CCSDS ASM patterns with length 128 and 192 bits have been obtained by a simple linear combination of shorter ASM patterns. The 128 bits sequence is the juxtaposition of the 64-bit sequence affirmed and complemented, represented in compact notation as ASM(128) = ASM(64) XOR (10). The same procedure has been applied to the 96-bit ASM for obtaining the 196-bit sequence via ASM(196) = ASM(96) XOR (10). (This property can be observed also by looking at the ASM binary code reported in Tab. 1). The autocorrelation functions for the 128 and 192 bits CCSDS ASM are shown in Fig. 3 and Fig. 4, and their secondary peak s are listed in Tab. 4 and Tab. 5, respectively.
Fig. 3. The autocorrelation function of CCSDS ASM(128) Tab. 4. of secondary peaks and s for CCSDS ASM(128) ±8 +14 ±64-64 V. THE NEW ASM PATTERNS Fig. 4. The autocorrelation function of CCSDS ASM(192) Tab. 5. of secondary peaks and s for CCSDS ASM(192) ±52, ±40 +20 ±96-96 The ASM patterns belonging to the CCSDS standard and presented in the previous chapter show good properties. Nevertheless new algorithms for designing binary sequences with good autocorrelation properties have been recently discovered and applied so that better sequences have been found. A. THE NEW ASMs WITH N=64 AND 96 BITS Two very good sequences for N=64 and N=96 have been recently obtained via natural evolution process algorithm [5]. This is a population-based optimization process that starts from a randomly chosen sequence and evolves using some decisions based on a fitness function. This algorithm slightly outperforms the results obtained in [4] presented in the previous section and adopted by CCSDS. The two new sequences are presented in Tab.6.
Tab. 6. The new Attached Sync Markers with length N = 64 and 96 bits Code-rate r ASM length N [bits] ASM (hex code) 1/2 64 8A4F774A 14640DC3 1/3 96 2B7EAFC1 2CD984AC E451CB43 The autocorrelation functions for the new 64 and 96 bits ASM are shown in Fig. 5 and Fig. 6, and their secondary peak s are listed in Tab. 7 and Tab. 8, respectively. Fig. 5. The autocorrelation function of the new ASM(64) Tab. 7. of secondary peaks and s for the new ASM(64) ±41, ±7 +5 ±29, ±25-5 Fig. 6. The autocorrelation function of the new ASM(96) Tab. 8. of secondary peaks and s for the new ASM(96) ±73, ±25 +7 ±81, ±79, ±1-7 In both cases there is a slight reduction of the peak side lobe level parameter with respect to the corresponding one found in the previous section.
B. THE NEW ASMs WITH N=128 AND 192 BITS The main problem of the longest CCSDS ASMs is that there is a high peak lobe for a shift corresponding to N/2, due to the method used for build them. A new algorithm for directly searching long sequences has been recently proposed in [6] The sequences obtained for N=128 and N=192 bit show excellent properties in terms of peak side lobe level. The two new sequences are presented in Tab. 9. Tab.9. The new Attached Sync Markers with length N = 128 and 192 bits Code-rate r ASM length N [bits] ASM (hex code) 1/4 128 507F9A45 63B7624D 634F3A0A 80A103C6 1/6 192 416CD506 19567B73 C6C1ADA1 AB47D47F 6AC7BE42 C9DCFE4A The autocorrelation functions for the new 128 and 192 bits ASM are shown in Fig. 7 and Fig. 8, and their secondary peak s are listed in Tab. 10 and Tab. 11, respectively. Fig. 7. The autocorrelation function of the new ASM(128) Tab. 10. of secondary peaks and s for the new ASM(128) ±110, ±72, ±64, ±32 +8 ±102, ±88, ±78, ±56, ±52-8 Fig. 8. The autocorrelation function of the new ASM(192) Tab. 11. of secondary peaks and s for the new ASM(192) ±131 +11 ±105, ±9-11
In both cases there is a very significant reduction of the peak side lobe level parameter with respect to the corresponding CCSDS ASM patterns described in the previous section. VI. ASM PATTERNS PERFORMANCE WITH NOISE AND QUANTIZATION The frame synchronizer computes the correlation between the local copy of the ASM pattern and the received noisy sequence; i.e. the ASM symbols surrounded by random data symbols, both of them corrupted by the noise. Additionally the frame synchronizer computes the autocorrelation function and compares it against a given threshold. If the computed is larger than the threshold, the ASM presence is assumed. Due to the stochastic nature of the received symbol sequence, two variables must be taken into account: False Lock: the frame synchronizer position is wrong but, due to the noise, the autocorrelation function exceeds the threshold. The probability of this event is called False Lock probability Missed Detection: the frame synchronization position is exact but, due to the noise, the autocorrelation function does not exceed the threshold. The probability of this event is called Missed ASM probability. The of these variables highly depends on the selected threshold: if the chosen threshold is low, the probability of False Lock increases, conversely if the threshold is high, the Missed ASM probability increases. For properly setting the threshold, a simulation campaign has been performed. First of all, we have chosen a target Frame Error Rate FER = 10e-4 as operating point for our simulations, which is typical for many space missions. The operating points for different input frame length k and code-rate r corresponding to this FER can be derived from the simulated performances reported in [2] and [3] and are listed in Tab.12. Tab. 12. Operating points for CCSDS Turbo codes k r E b /N 0 [db] E r b /N 0 [db] E r b /N 0 [db] E r b /N 0 [db] FER=1e-4 FER=1e-4 FER=1e-4 FER=1e-4 1784 1/2 1.47 1/3 0.80 1/4 0.59 1/6 0.29 3568 1/2 1.25 1/3 0.60 1/4 0.37 1/6 0.09 7136 1/2 1.10 1/3 0.45 1/4 0.22 1/6-0.05 8920 1/2 1.05 1/3 0.43 1/4 0.20 1/6-0.10 Then, we have also considered the effect of quantization. The samples of the received symbols are processed and stored using a fixed number of quantization bits. In particular, in this work we used 8 and 3 bits for quantization process, which correspond to the s actually used in ESA's receivers. Therefore, the optimum threshold depends on its sensitiveness of the False Lock and the Missed ASM probabilities and to the given constraints. For each Eb/N0 considered we have evaluated (via a very large number of simulations) the false lock and missed detection probability as a function of the threshold. This way, given an ASM, we have individuated the best threshold for different operating points. The observed improvements are rather modest for the two shortest lengths 64 and 96 bits because the corresponding CCSDS sequences are already very good, therefore the new ASMs are just slightly better. Referring to the two longest pattern lengths, the improvements are extremely high because the CCSDS ASMs with length 128 and 192 bits suffer from the presence of high PSL ratio that is responsible of performance's lowering. As a significant example, the ROC performance for the CCSDS and the new ASMs with length 128 and 192 are reported in Fig. 9 and Fig. 10. The strong improvement obtained by the new markers is remarkable. It is interesting to note that, given an instantaneous E b /N 0 measure from the demodulator, the threshold can be tuned according to this, allowing optimum performance at any moment. As an example, in a typical scenario characterized by low E b /N 0, careful threshold tuning allows the system not to lose synchronization. The system would remain locked-in as well, no matter if turbo decoder is not able to work properly for a while. It is important to consider which impact the integration of new ASM bit patterns into the standard [1] would have. While receivers are normally configurable and/or easily modifiable, the spacecraft transmitters rely more on less flexible hardware implementations. This issue is likely to impact any proposed modification since compatibility with existing ASIC implementations shall be preserved,
Fig. 9. ROC curves for the CCSDS ASM (128) and the new ASM(128), with 8-bit quantization Fig. 10. ROC curves for the CCSDS ASM(192) and the new ASM(192), with 8-bit quantization VIII. CONCLUSIONS In this paper we investigated the performance of recently found binary sequences with good properties in terms of aperiodic autocorrelation function. When compared against the current CCSDS ASM patterns, the new sequences overshoot the current ones since they superior autocorrelation properties lead to better False lock and Missed ASM performance in real scenarios. They ensure a considerable gain in terms of link budget that could be used to improve overall performances or to relax some constraints on devices. Despite the new sequences could simply replace CCSDS ones with no modification on the synchronization algorithms, configurability of such synchronization algorithms is required in order to preserve compatibility with existing ASICs. This may lead to the addition of the new patterns instead of replacement of the current ones.. REFERENCES [1] CCSDS 131.0-B-1, TM Synchronization and Channel Coding, Blue Book, Issue 1, September 2003. [2] CCSDS 130.1-G-1, TM synchronization and channel decoding - summary of concept and rationale, Green book, June 2006. [3] Gian Paolo Calzolari, Franco Chiaraluce, Roberto Garello, and Enrico Vassallo. Turbo Code Applications: a journey from a paper to realization, Chapter 13: "Turbo Code Applications on Telemetry and Deep Space Communications", Springer-Verlag, 2005, pp. 321-344. [4] F. Hu, P.Z. Fan, M. Darnell, and F. Jin. Binary sequences with good aperiodic autocorrelation functions obtained by neural network search. Electronic Letters, September 1997, pp.688-690. [5] X. Deng and P. Fan. New binary sequences with good aperiodic autocorrelation functions obtained by evolutionary algorithm. IEEE Communication Letters, October 1999. [6] Roberto Garello, Noureddine Boujnah, and Yifan Jia, Design of Binary Sequences and Matrices for Space Applications, Proceedings of 2009 International Workshop on Satellite and Space Communications - IWSSC 2009, Siena, Italy, September 2009.