A complete Orthogonal Frequency Division Multiplexing (OFDM) synchronization chain is proposed based on practical synchronization schemes. The synchronization is accomplished in three steps: frame detection (or timing), frequency offset estimation, and phase tracking. OFDM is one of the up and coming multiplexing techniques that have been proposed to be used in 4th Generation Wireless systems. The basic principal of OFDM is crucial as it is gaining widespread popularity in the wireless community.
Two algorithms are described for performing frame (and OFDM symbol) timing acquisition and frequency-offset estimation in Fixed Broadband Wireless Access OFDM systems. They are evaluated in two different radio frequency bands, namely in 5.8 GHz where a severe multi-path environment is expected under Non-Line-Of-Sight (NLOS) conditions, and in the licensed 10.5 GHz band, under Line-Of-Sight (LOS) conditions. In addition, a new preamble structure with duration of two OFDM symbols is proposed here, that actually gives rise to the techniques proposed herein. The first algorithm uses only the 1st symbol and gives an averaging enhancement with minimal overhead complexity to a well-known correlation-based technique. The second algorithm uses the 1st symbol for coarse and the 2nd symbol for fine symbol timing synchronization. It is proven that both algorithms significantly reduce the start of frame acquisition error in a wide range of SNR values.
Introduction
Orthogonal Frequency Division Multiplexing (OFDM) has gained considerable interest in recent years. Orthogonal Frequency Division Multiplexing (OFDM) is an attractive modulation method for channels with a nonflat frequency response, as it saves the need for complex equalisers and where data symbols are transmitted in parallel by employing a number of orthogonal sub-carriers. A block of N serial data symbols, each of duration Ts, is converted into a block of N parallel data symbols, each with duration T=NTs. The N parallel data symbols modulate N sub-carriers that are spaced 1/T Hz apart. One of the biggest problems of OFDM is the synchronization. To ensure ISI-free detection, precise timing information (regarding where the symbol boundary lies) is needed so that an uncorrupted portion of the received OFDM symbol can be sampled for FFT. To ensure subcarrier orthogonality and hence ICI-free detection, the transmitter receiver carrier frequency offset must be estimated and compensated. There has been a body of literature devoted to OFDM synchronization. Almost all schemes are based on correlation of repeated signal. An earlier work by Moose [2] uses combined short and long pilot symbols for frequency-offset estimation. Van De Beek described maximum likelihood timing and frequency offset estimator using cyclic prefix in. A pilot-based timing and frequency offset estimation scheme is proposed by Schmidl in [4] that uses a pilot symbol with half of carriers modulated (effectively dividing one OFDM symbol into two identical halves). In this paper we propose a complete OFDM digital synchronization chain as opposed to most of the papers to date, which only addressed certain parts of the OFDM synchronization. A number of new schemes are also proposed to implement the chain including: differential PN sequence for precise frame detection, long OFDM pilot symbol with PN sequence differentially coded across sub-Carriers for frequency offset estimation, and digital phase locked loop to correct frequency offset estimation residual error and phase noise.In order to successfully deploying a Fixed Broadband Wireless Access system, one has to employ a transmission technique that yields low bit error rate over frequency-selective fading channels. OFDM is a technique that counters such channel environments with reasonable complexity. However, there are a couple of technical challenges to solve. First, OFDM is extremely sensitive to carrier frequency offsets, which are mainly caused by the inherent instabilities of the carrier frequency oscillators in the transmitter and the receiver. To employ OFDM, these offsets must be very small compared to the subcarrier spacing, which is NFFT-times smaller than the bandwidth of the original clock rate (NFFT is the number of OFDM subcarriers); otherwise, the carrier frequency offsets force the orthogonality between the subcarriers to be lost, which severely degrades the performance. In addition, symbol timing synchronization must be achieved within an acceptable preamble period. In OFDM, symbol-timing errors may cause inter-symbol interference (ISI), since FFT windows can include adjacent OFDM symbol components. This also leads to inter-channel interference (ICI) that also causes essential degradation in channel estimates needed by coherent systems [4].
In a continual transmission system such as DVB-T, the requirements for the accuracy of the channel estimate and for frame-error-rate are very high. Since synchronization acquisition time is not so critical in this context, symbol-timing estimate may be improved by averaging over several symbols a correlation based metric that uses the cyclic prefix. Averaging is important since the precision of individual estimates is not satisfactory, even if long guard intervals reduce this averaging need. However, when it comes to jointly estimate the frequency offset, a cyclic prefix based time domain correlation method has a maximum detectable frequency error equal to half the subcarrier spacing f...
This can be increased if the spacing between the sample pairs used for the correlator's output is smaller. If the correlator uses identical adjacent symbol partitions of length L, then the maximum detectable frequency error increases to
In this work, two synchronization algorithms are proposed and evaluated and a new preamble with special properties is designed, consisting of two parts (part A & part B) of 1 OFDM symbol duration each. The first proposed algorithm takes advantage of the special structure of part A in order to give an averaging enhancement with minimal overhead complexity to the correlation-based technique presented in The second proposed algorithm combines the technique in for a coarse (using part A) and the algorithm in for fine symbol timing estimate (using part B).
Section 2 defines the system and channel models used and the proposed preamble structure. In Section 3 a description of the two algorithms is given and simulation results are presented, while Section 4 concludes the work presented herein.
System and channel models
This work was performed under the framework of ADAMAS IST project. The objective of ADAMAS is to demonstrate a dynamic TDD/TDMA fixed broadband wireless access system reaching up to 20 Mbps aggregate user throughput in two frequencies, namely in 5.8 GHz (unlicensed band) and in 10.5 GHz (licensed band). The system is OFDM based with an FFT size of 256 and the channel bandwidth is 7 MHz for both frequencies.
One channel model (channel-A at 5.8 and channel-A at 10.5 GHz) has been used for the simulations for each of the two frequencies, representing typical cases for the two environments. These models were derived by the channel modelling activity of ADAMAS. Channel coherence time in these environments is expected to be high, thus the channels are considered static for the duration of each data burst (i.e., for several OFDM symbols). The cyclic prefix consists of 48 samples for both systems. Channel-A at 5.8 GHz represents a very severe environment where the multi-path effects are dominating due to NLOS conditions. On the other hand, channel-A at 10.5 GHz has a very strong direct path and was modelled with Rician distribution.
A new preamble structure has been designed in order to make an efficient usage of the synchronization algorithms presented herein. The general preamble structure preceding the data OFDM symbols is illustrated in the following Figure:
 | |
| Figure 1: General preamble structure |
Figure 1: General preamble structure
 |
| Figure 2: Part A of the preamble |
It consists of 6 short sections: 4 identical short sections of type a with length of 32 samples each, and 2 identical short sections of type b of length 64 samples each. The cyclic prefix (CP) is a copy of the 48 last samples. Part B of the preamble is designed by BPSK modulation of all the used subcarriers using a PN sequence with excellent auto-correlation properties; the receiver can use it also for channel estimation and it has no identical parts.
Description and evaluation of synchronization techniques
Two preamble-based techniques were the basis of the work presented in this paper. In this Section an overview of the proposed techniques and as well as the modifications of them are given. The whole work was based on the new preamble structure. The relevant simulation results are also given herein.
The basic principle of synchronization is the correlation between two identical parts of the preamble. T. Schmidl and D. Cox in proposed a synchronization method using a training symbol consisting of two identical halves in time, each of length N, and it is used for frame and symbol timing synchronization and frequency offset estimation. The algorithm attains the correlation-based metric by correlating the conjugate of the first half symbol with the second half. An improvement in the performance is possible with the use of a weighting function, as proposed in .
Therefore:
Refer section{ A} below
Refer Section {b}
Phase rotation caused by the carrier frequency offset is obtained by getting the angle of the peak value P (d) the correlator’s output, from which we can easily derive the estimate of the frequency offset between the TX and RX oscillators. The special nature of part A of the preamble we propose in this work gives rise to an averaging enhancement in the detection of the maximum M (d) within one OFDM symbol duration. This can be easily shown by giving an example of how the M (d) function looks like when it is applied to part A of the preamble. The following figure depicts M (d) under 10 dB SNR, using channel A at 10.5 GHz.
Each of the two big distinct peaks are detected when the window of the correlator (of length 2*N=64 in our case) is aligned with the two short a of length 3 samples. Since there exist two pairs of short a , and we know that their relative position is expected to be 128 samples apart (see Figure 2) we can average the actual two detected positions and get a better estimate for the maximum M (d) and its position. This leads to a better start of symbol acquisition and frequency offset estimation, as will be shown in the simulations that follow. Since the detection of peak existence is commonly implemented by inspecting if the level of the correlator's output becomes bigger than a predefined threshold, the existence of two peaks decreases the possibility of no-peak detection at all. By inspecting Figure 3, we also see that there is no plateau near the maximization points of M (d),which a phenomenon that arises when the cyclic prefix is actually a part of the identical parts that are being correlated. As it is also mentioned in, this plateau affects timing synchronization, since it is difficult to detect the peak, however the frequency-offset estimation is robust against timing uncertainties. In our case, the prefix in part A of the preamble (Section 2) is uncorrelated to the identical parts that are being correlated and this plateau does not exist. Also, by using the a sections having their equal samples being 32 time samples apart, we have a detectable frequency offset of about 6 and 11 ppm at 10.5 and 5.8 GHz respectively.
Simulation results for the 1st technique (use of part A of the preamble)
The averaging technique is compared to the conventional technique that detects only one (i.e., the 1st) peak of the correlator's output. This comparison is carried out in terms of pdf (probability density function) of the observations taken. Simulations were derived for SNR values of 6, and 20 dB. In all cases, a frequency offset of 100 kHz was injected in the system. The slope a w of the weighting function was set to 0.
Figures 4 to Figure 7 depict the pdfs of the results taken for symbol start acquisition, under channel A at 10.5 GHz, using the conventional (Figures 4,6) and the averaging (Figures 5,7) methods for the 2 different SNR values. Results located at point 0 indicate perfect acquisition. Results that lay at a positive number indicate that this number of samples delayed the acquisition, thus the useful OFDM symbol obtained will have also samples from the next OFDM symbol, which will cause ISI. Results that lay at a negative number indicate that the acquisition actually started within the cyclic prefix of the same symbol. If this location is within the ISI free region of the cyclic prefix, only an easily correctable (by the equalizer) linear phase shift in the subcarriers will take place. One can easily see that by employing the averaging algorithm we get a better concentration of the results towards the actual correct synchronization point (sample 0). Under 6 dB SNR, a small number of observations were found at sample 20 (and beyond), which means that some peaks were not detected, and this is the reason for the high std value. This can be avoided by slightly decreasing the “start of a peak” detection threshold; here it was set to 0.45. depict the pdfs of the results taken for frequency offset estimation error, as a percentage of sub-carrier spacing. What we can derive from these Figures is that the averaging method is slightly better than the conventional. The rather small improvement can be justified since frequency offset estimate is robust against timing uncertainties. The residual frequency offset in the data OFDM symbols can be corrected with a pilot assisted technique in the frequency domain.
0 comments:
Post a Comment
Thanks for your Valuable comment