CN1277359C - Carrier Frequency Offset Estimation Method for Orthogonal Frequency Division Multiplexing Communication System - Google Patents

Abstract

The present invention relates to a method for estimating the carrier frequency shift of an orthogonal frequency division multiplex (OFDM) communication system. Two kinds of training sequences designed specifically are added to initial positions of sequence frames, cross correlation results of the A kind of the training sequences are utilized to estimate the original value of frequency shift, namely the rough estimation of the frequency shift, and then, the B kind of the training sequences carries out autocorrelation so as to obtain the accurate estimation of signal frequency deviation within a small range, namely the accurate estimation of the frequency shift; then, the two estimated values are combined so as to obtain the exact value of system frequency shift; finally, the digitally controlled oscillator is controlled to produce the rectifying value of the system frequency shift, and the rectifying value is multiplied by time domain data so as to carry out the compensation of the frequency shift. With the advantages of high accuracy, small amount of calculation and easy realization for a system, the method of the present invention can be used in an OFDM wireless local network system and can also be used in the process of realizing various OFDM systems.

Description

Carrier Frequency Offset Estimation Method for Orthogonal Frequency Division Multiplexing Communication System

Technical field:

The invention relates to a method for estimating carrier frequency offset in an orthogonal frequency division multiplexing communication system, which is a method for estimating and compensating carrier frequency offset, and belongs to the field of digital communication.

Background technique:

Orthogonal Frequency Division Multiplexing (OFDM) is a data modulation method that converts high-speed serial data into low-speed parallel data, and modulates the parallel data with multiple mutually orthogonal subcarriers. Although sub-carrier spectrum overlaps, since each sub-carrier is orthogonal, there is no interference between carriers, so spectrum utilization efficiency can be effectively improved. As an emerging multi-carrier modulation technology, OFDM has strong anti-multipath fading and frequency selective fading characteristics, and has been used in asymmetric digital subscriber line (ADSL), digital television broadcasting (DVB), wireless local area network (IEEE802.11a ) and many other fields have been widely used.

In order to ensure the orthogonality among the subcarriers of the OFDM system, the frequency of the subcarriers must satisfy: f _n =f ₀ +n/T, where T is the symbol period. Therefore, the frequency interval Δf=1/T between the subcarriers. Due to the parameter drift of system components and the influence of Doppler frequency shift, the carrier frequency at the receiving end will shift, which will lead to serious inter-carrier interference (ICI). If the frequency offset error is an integer multiple n of the subcarrier frequency interval Δf, the received signal is shifted by n subcarriers in the frequency domain. At this time, the subcarriers still remain orthogonal, but it will completely destroy the correct reception of data. If the frequency error is a fractional multiple of the subcarrier interval, it will cause the energy of other subcarriers to be aliased at the sampling point in the frequency domain, that is, cause ICI, thereby affecting the BER performance of the system. Therefore, the carrier frequency synchronization must be performed at the receiving end, that is, the carrier offset error is estimated and compensated.

Generally, frequency synchronization is performed after the initial correction of the time synchronization error. According to the position of the estimation algorithm in the system, it can be divided into two types: time domain frequency synchronization and frequency domain frequency synchronization. The frequency domain synchronization algorithm was proposed by Ferdinard Classen and Heinrich Meyr in the paper Frequency Synchronization Algorithms for OFDM Systems Suitable for Communication Over Frequency Selective Fading Channels (in Proc.IEEE Veh.Technol.Conf., 1994, pp.1655-1659). It is a method that uses repeated sequences combined with DFT to estimate the frequency error. It can improve the estimation range of the frequency offset, but it requires the pre-processing of the FFT data. These processes will actually cause additional errors, resulting in actual processing delays, and also Increased the difficulty of system implementation. The time-domain frequency synchronization method was proposed by Paul HMoose in the paper A Technique for Orthogonal Frequency Division Multiplexing Frequency Offset Correction (IEEE Trans. Commun., vol.42, pp.2908-2914, Oct.1994), which is mainly based on the maximum likelihood Although (MLE, Maximum LikelihoodEstimation) frequency offset estimation algorithm, but it has a large amount of calculation, low precision, and is not easy to implement in the system ASIC.

Invention content:

The purpose of the present invention is to address the deficiencies in the prior art, and propose a method for estimating carrier frequency offset in an OFDM communication system. Requirements for frequency offset estimation range.

In order to achieve this goal, in the technical solution of the present invention, firstly, at the initial position of the system frame, a specific training sequence is added to facilitate the estimation of the frequency offset; secondly, the cross-correlation results of the added part of the training sequence are used to estimate The initial value of the frequency offset is to perform a rough estimation of the frequency offset; then use the added part of the training sequence to perform autocorrelation to obtain an accurate estimate of the signal frequency offset in a small range, that is, to perform a fine estimation of the frequency offset; then the two The value of the second estimate is combined to obtain the precise range of the system frequency offset; finally, the numerically controlled oscillator (NCO) is controlled to generate the correction value of the system frequency offset, and it is multiplied with the time domain data to compensate for the frequency offset. Concrete operation of the present invention is carried out as follows:

1. Add a specially designed training sequence at the initial position of the system frame.

Due to the need of frequency offset estimation, two types of training sequences are added before the data to be sent. Among them, the number of training sequences in the former type is more, but the sampling points of a single sequence are less, which is called type A training sequence. Immediately after the A-type training sequence is another training sequence with a small number but a large number of sampling points in a single sequence, which is called a B-type training sequence. The cyclic prefix (CP) of the B-type training sequence is a copy of several data at the end of the B-type training sequence. After the A and B training sequences, there is data to be sent in system frames. The number of sampling points of a single type A training sequence is  of the number of sampling points of data to be sent in system symbols.

In this way, the class A training sequence of each frame can be used for rough frequency synchronization or rough estimation, and then the cyclic prefix of the class B training sequence and the cyclic characteristics of the data can be used for fine frequency synchronization or fine estimation.

2. Use the A training sequence to roughly estimate the frequency offset.

Use the last several fully received training sequences of type A to perform cross-correlation, and estimate the initial value of the frequency offset according to the result, that is, perform a rough estimation of the frequency offset.

The expression of the rough estimated value f _δ1 of the frequency offset error is

Among them, N is the length of the sampling point of the data to be sent in the system frame, L _a is the length of the sampling point of a single type A training sequence, Δf is the frequency interval between each subcarrier of the system,

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; AEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; AEnd</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>$

Among them, z(k _AEnd ) is the last time-domain sampling point of the last A training sequence, where the received data of the length of the two A-type training sequences are correlated to play the role of numerical average.

Since N=4L _a , the roughly estimated range of f _δ1 is [-2Δf, 2Δf].

3. Using the B training sequence to perform fine estimation of the frequency offset.

Using several completely received training sequences of type B to perform autocorrelation with the cyclic prefix of the training sequence of type B, and estimate the precise value of the frequency offset according to the result, that is, perform fine estimation of the frequency offset.

The expression of the frequency offset fine estimated value f _δ2 is

It can be seen that the estimated range of the fine estimate is reduced to [-Δf/4, Δf/4]. in,

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; cpb</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; BEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; BEnd</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>$

Where z(k _BEnd ) is the last time domain sampling point of the last class B training sequence, N _b is the number of B class training sequences, L _b is the sampling point length of a single B class training sequence, and L _cpb is the B class training sequence The sample point length of the cyclic prefix of the sequence.

4. Combining the rough frequency estimation and the fine frequency estimation to obtain a frequency offset value.

Combining the frequency rough estimation and fine estimation, the frequency offset value f _δ of the system is expressed by the following formula:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\frac{\mathrm{<span\; class="notranslate">\; \&Delta;\; f</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; \&delta;</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}$

(

为下取整符号)，也即可以利用f_δ1和f_δ2得到频率偏移的精确的估计值。这种算法的频率偏移估计总的范围为[-2Δf，2Δf]。in

(

is the lower integer sign), that is, f _δ1 and f _δ2 can be used to obtain an accurate estimate of the frequency offset. The frequency offset estimation of this algorithm has a total range of [-2Δf, 2Δf].

5. Use NCO (Numerical Controlled Oscillator) to compensate for frequency offset.

After the frequency offset value is calculated, a numerically controlled oscillator (NCO) can be used to correct the frequency offset, that is, the time domain data is multiplied by the corrected value of the frequency offset: e ^-j2πfδnTs , where T _s is the system sampling interval. The frequency offset compensation is performed after the frequency fine estimation and rough estimation are completed and their results are integrated, and the correction value is obtained according to the estimation result and the NCO is controlled to generate the frequency offset correction value and multiply it with the time domain signal.

The present invention firstly obtains the approximate range of the frequency offset through rough frequency estimation, and further refines the range of frequency offset through fine frequency estimation, so the precision of frequency offset estimation is high. The frequency offset estimation algorithm of the present invention is implemented in a wireless local area network, and the obtained results show that the measured frequency error is relatively small, which proves that the present invention has an obvious effect on estimating the frequency offset of the system, so that the receiving end can better recover the user data. Compared with the frequency domain frequency synchronization algorithm, the method of the present invention does not need to wait for the pre-processing process of the FFT data, and the system time delay is small; compared with the frequency offset estimation algorithm based on maximum likelihood, the method of the present invention has a small amount of calculation, High-precision, easy-to-system ASIC implementation. The method of the present invention not only has a good effect on frequency offset estimation, but also is easy to implement. Therefore, the method for estimating carrier offset based on a specific training sequence is an ideal method for estimating system frequency offset.

The present invention can be used in an OFDM wireless local area network system, and can also be applied in the realization process of OFDM systems such as DVBT, CDMA-OFDM and OFDMA.

Description of drawings:

Fig. 1 is a schematic diagram of the OFDM system frame structure after inserting the training sequence designed by the present invention.

As shown in FIG. 1 , the frame structure includes two types of training sequences, A and B, and data to be sent, wherein both the training sequence of type B and the data to be sent include a cyclic prefix (CP). The number N _a of type A training sequences is greater than the number N _b of type B training sequences, but the number of sampling points of a single type A sequence is smaller than the number of sampling points of type B training sequences. The cyclic prefix (CP) of the B-type training data is a copy of several lengths of data at the end of the B-type training sequence. After the inserted A and B training sequences are the original data to be sent in the system frame.

Fig. 2 is the implementation structure of the frequency synchronization algorithm of the present invention.

As shown in Fig. 2, the method of the present invention includes several parts of frequency rough estimation, frequency fine estimation, frequency estimation synthesis and frequency compensation by using NCO. Among them, the frequency is roughly estimated by multiplying and summing a class A training sequence and the preceding class A training sequence obtained after delay, that is, obtaining the phase angle of the result after cross-correlation to obtain the result of rough synchronization. The frequency fine estimation is obtained by multiplying and summing a class B training sequence with the cyclic prefix of the preceding class B training sequence obtained after delay, that is, the phase angle of the result obtained after autocorrelation is obtained to obtain a fine synchronization result. The results of these two steps are synthesized to obtain the frequency control word FCW to control the numerically controlled oscillator NCO, and the value generated by the numerically controlled oscillator is multiplied by the time domain data to compensate for the frequency offset.

K in the figure is to ensure that the first value of the B-type training sequence CP is input for frequency fine estimation. In each clock cycle of the NCO, the phase accumulator increases sequentially with the value of the frequency control word, and the calculated phase value is sent to the sine and cosine generators, and then multiplied by the system baseband signal to compensate for the frequency offset.

Fig. 3 is a schematic diagram of the application of the present invention in an OFDM wireless local area network transmitting and receiving system.

As shown in Figure 3, at the transmitter, the data to be sent is firstly subjected to convolutional encoding, interleaving, and modulation, and then two specially designed training sequences, A and B, are added before the data to be sent to estimate the frequency offset. After the data has been processed, N-point IFFT is performed, the cyclic prefix of the data part is added, and modulated to the intermediate frequency and radio frequency. At the receiving end, after the intermediate frequency and radio frequency data are demodulated into baseband signals, time synchronization is performed first, and then frequency offset estimation and compensation work is performed. Among them, the frequency offset estimation is to first use the A-type training sequence to perform a rough estimation of the frequency offset, and then use the B-type training sequence to perform a fine estimation of the frequency offset, and combine the results of the rough and fine estimation to obtain an accurate frequency offset value and control the generation of the NCO. The frequency offset correction value is multiplied with the system baseband signal to complete the frequency offset compensation. In this way, the estimation and compensation work of the system carrier offset is completed. Then the data can be restored to user data after removing the cyclic prefix, N-point FFT, frequency equalization, demodulation, and decoding.

Detailed ways:

The solution of the present invention will be further described below through specific examples.

Embodiment: The method of the present invention is applied to OFDM wireless local area network.

The application of the present invention in an OFDM wireless local area network transmitting and receiving system is shown in FIG. 3 . After the data at the transmitter is convolutionally encoded, interleaved, and modulated, a specially designed training sequence is first added to the system frame to estimate the frequency offset. Here, since the data sampling point length to be sent in the system frame is 64, the sampling point length of each A training sequence added is $\frac{1}{4}\&times;64=16,$ There are 10 training sequences for class A. According to the relationship between the lengths of training sequences A and B, two training sequences of type B are added, the length of sampling points of each training sequence of type B is 64, and the cyclic prefix of training sequences of type B is 32. Perform 64-point IFFT on the improved frame (that is, the modulation process of OFDM), add the cyclic prefix CP of the data to be sent, and modulate to the intermediate frequency and radio frequency.

At the receiving end, after the data is demodulated into a baseband signal, time synchronization is performed first, and then frequency offset estimation and compensation work is performed. The frequency offset estimation is to use the added A-type training sequence to perform a rough estimation of the frequency offset, and then take the last two fully received A-type training sequences for cross-correlation for rough estimation. The formula for the rough estimation is:

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 15</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; AEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; AEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>$

Taking the phase of R _A , the expression of the rough estimation value f _δ1 of the frequency offset error can be obtained,

The range of rough estimate f _δ1 can be obtained as [-2Δf, 2Δf]

Then use the B-type training sequence to perform fine estimation of the frequency offset, and perform autocorrelation between the B-type training sequence and its cyclic prefix for fine estimation. The formula for the fine estimation is:

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 31</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; BEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; BEnd</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 128</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>$

Taking the phase angle of _RB , the expression of the precise estimated value of frequency offset f _δ2 can be obtained:

The estimated range in which the fine estimate can be obtained is [-Δf/4, Δf/4].

The results of rough and fine estimation are integrated to obtain an accurate frequency offset value, and the NCO is controlled to generate a frequency offset correction value, and the system baseband signal is multiplied to complete frequency offset compensation. After the data is processed, the user data can be obtained after de-cyclic prefix, N-point FFT, frequency equalization, demodulation, and decoding.

The present invention has achieved obvious effects in the implementation of the wireless local area network environment. When the signal-to-noise ratio is between 0-45dB, the maximum value of the measured carrier offset error is only 2.7kHZ, which does not exceed 1% of the subcarrier frequency interval in the wireless local area network. . In practical systems, the resulting ICI is already completely negligible. Therefore, adopting the method of the present invention can better overcome the system carrier frequency offset and eliminate the bad influence of the system carrier frequency offset.

The method of the present invention can not only estimate and compensate the carrier offset of the OFDM wireless communication system, including such as 802.11 series protocols, but also can be applied to OFDM transmission systems such as CDMA-OFDM, OFDMA, and DVBT. The difference is only the difference between the frame structure and the overall structure of the system. In the 802.11 protocol, the modulation method of OFDM is adopted; while in the CDMA-OFDM system, the combination of spread spectrum and OFDM is adopted, but the method of the present invention is applied in these systems. The method is basically the same.

Claims (2)

1. A carrier frequency offset estimation method of an orthogonal frequency division multiplexing communication system is characterized by comprising the following steps:

1) adding two types of specially designed training sequences before each frame of data, wherein the number of A type training sequences is large, the number of single sequence sampling points is small, the number of B type training sequences is small immediately after the A type training sequences, the number of single sequence sampling points is large, and a cyclic prefix CP of the B type training sequences is a copy of a plurality of length data at the tail of the B type training sequences;

2) multiplying and summing a certain class A training sequence and the class A training sequence which is completely received in a delayed way, namely performing cross correlation, and estimating an initial value of frequency offset according to a phase angle of a result, namely performing rough estimation of the frequency offset;

3) multiplying and summing a certain B-type training sequence and the cyclic prefix of the B-type training sequence which is completely received before the B-type training sequence after time delay, namely performing autocorrelation, and estimating an accurate value of frequency offset according to the phase angle of the result, namely performing fine estimation of the frequency offset;

4) integrating the values of the fine frequency offset estimation and the coarse frequency offset estimation to obtain the accurate range of the system frequency offset;

5) and controlling the NCO to generate an estimated correction value of the system frequency offset, and carrying out complex multiplication on the estimated correction value and the system baseband signal so as to carry out frequency offset compensation.

2. The method of claim 1, wherein the number of samples of a single class-a training sequence is * times the number of samples of the data to be transmitted in the systematic symbols.

Images