GB2388754A

GB2388754A - Frequency burst error estimation

Info

Publication number: GB2388754A
Application number: GB0210903A
Authority: GB
Inventors: Ma Ming
Original assignee: Matsushita Electric Industrial Co Ltd
Current assignee: Panasonic Holdings Corp
Priority date: 2002-05-13
Filing date: 2002-05-13
Publication date: 2003-11-19
Anticipated expiration: 2022-05-13
Also published as: GB2388754B; GB0210903D0

Abstract

An improved method for frequency burst error estimation in which the centre complex correlation value used to derive the accumulated phase error estimation is calculated from the expression <EMI ID=2.1 HE=12 WI=37 LX=569 LY=704 TI=MF> <PC>where <EMI ID=2.2 HE=14 WI=29 LX=423 LY=837 TI=MF> <PC>fs = sampling frequency and W fn = initial frequency offset of receiver. The method can be applied in TDMA systems, e.g. GSM.

Description

', 2388754

Frequency Burst Error Estimation The present invention relates to a method and apparatus for detecting a frequency burst signal in a received digital signal and more particularly to estimating actors in phase and hence frequency of received signals. The invention has been developed for use in detecting frequency burst signals in received digital telecommunication signal frames but it will be appreciated that the invention may have other applications.

It is well known that accurate reception and transmission of data in mobile telephone systems depends on proper synchronization of mobile stations with base stations. One method of enabling such synchronization to be achieved involves broadcasting frequency bursts (FB) on a so-called broadcast common channel. A frequency burst consists of 3 tail bits, 142 fixed data bits with state "0", 3 further tail bits and 8.25 guard bits. The 142 fixed data bits with state "0" correspond to an unmodulated carrier 67.7 kHz above the normal carrier frequency in a TDMA (time division multiple access) time slot if the mobile terminal is synchronized in frequency with the base station. However, before this frequency synchronization has taken place, the received FB frequency will be in the range of 67.7kHz + 11.5kHz for the GSM system in 900MHz band or in the range of 67.7 + 23kHz for the GSM system in 1800MHz band due to the inaccuracy of the mobile terminal reference oscillator, which is about + 12ppm. The carrier 67.7 kHz is equal Fed, where Fs is the sampling frequency, 270,833 Hz. The algorithm for detecting the frequency burst is briefly described below. The principles of the algorithm are discussed in more detail in GB-A-

23 15198.

After the reception of complex signal samples hen), a DC offset compensation function is usually carried out to remove the do part of the input samples. The DC offset compensation is taken every one burst (156 bits):

average =-at, S(i) S(i) = S(i) - average (2) The DC offset compensation is followed by the normalization function which is carried out every 39 (156/4) samples: S(i) = a;;; (3) A window complex correlation is then computed according to the following equation: j=N Corr(i) - S(i + j + M) x S(i + j) (4) J=o where the asterisk denotes complex conjugation, N is the correlation computing window length, the largest value of N is 141, for implementation convenience, Nis selected as multiples of 16. M ranges in value from 3 to 7 (typical value 5 or 6) which takes the GMSK modulation and multipath propagation into account. The value of the complex correlation function Corr(i) needs to be optimised in the FB detector test. If there is no FB present, the magnitude of the above correlation value is very small. If a certain threshold is exceeded for certain search window distance, the FB detector declares a FB is present. This threshold needs to be optimized in the FB detector test.

The value produced by the correlation calculation will increase to a maximum when all 142 samples (assuming N = 141) are within the frequency burst. The index i with maximum correlation magnitude is the estimated start position of the FB signals.

i In practice, to avoid repeatedly computing and adding correlation values following the equation 1, each new correlation value can be calculated by subtracting the earliest samples from the previous correlation value, adding the latest samples to the previous correlation value according to the following equation: Corr(i + 1) = C.orr(i) S(i + M) * S(i) + S(i + + M + 1) * S(' + N + 1) (5) When the start position of the FB signals has been found, the centre K samples of the FB are taken through a digital band pass filter which has the characteristic of a centre frequency at one quarter of the samples frequency and a bandwidth of 46kHz (corresponding to the case of frequency offset i23kHz). The selection of K is dependent on the allowed TOA (time of arrival) estimate error. For example, if the allowed TOA estimate error is _ 20 bits, K can be 142-20*2 = 102. The actual value of K needs to be optimised based on the frequency offset estimate error. The K output samples Sbp^) of the band pass filter are used to compute the correlation value following equation: À =K AP=>,Sbpf (i+l +TOA+71-K/2,iShpf (i+TOA+71 -K/2) (6) To where AP is a complex value corresponding to accumulated phase over the K samples.

For the GSM FB signals, the successive samples are rotated by /2. Thus the arctangent computation of the AP will obtain a value 12 + A<p, where A<pis an accumulated phase error. The frequency offset estimate can be obtained by Of = $ S' (7) where As is the sampling frequency, equals to 270.833kHz.

In practical implementation, because there is a low pass filter with cutoff frequency of 96kHz ahead of FB detector, only a high pass filter with cut-off frequency of 47 7kHz is needed instead of a band pass filter. This high pass filter is not only for frequency offset estimation, but also can be used to remove DC offset of the input samples. This high pass filter will be implemented as a digital Chebyshev type II filter which order should be made as low as possible for implementation reason.

According to the present invention, as defined in claim 1, the value " 1 " in equation (6) above can be replaced by a larger integer. The use of a larger integer can reduce fluctuations in the frequency offset estimation due to noise and fading. In order to avoid ambiguity of phase, it is preferred that the larger integer, J. obey the equation.

2Afn whereas= sampling frequency and AN = worst case frequency offset of receiver.

An embodiment of the invention will now be described by way of example only and with reference to the accompanying drawing in which: Figure is a histogram comparing the frequency offset estimate performance using a conventional algorithm and the performance using an algorithm according to the invention; and Figure 2 is a graph comparing root mean square error performance using a conventional algorithm and an algorithm according to the invention.

Modifying equation (6)according to the invention gives the following:

=x AP=Shp, (it J+TOA+71-K/2hpi(i+TOA+71-K/2). (8) =0 As noted above, this algorithm can potentially reduce the fluctuations occurring in the frequency offset estimation due to noise and fading. If J= 4, then arctangent computation of the AP will be a value of damp, if J = 5, then arctangent computation of the AP will be a value of JO + damp. Following the equation (7), the frequency offset estimate can be obtained. It is found that this modified algorithm gives much better frequency estimation performance than the conventional one as shown in Figures I and 2.

J can be any integer greater than 1. For practical purposes there is an upper limit to J to avoid ambiguity of phase defined by the expression: 1 * 1 < (9)

Combining equations (7) and (9) results in the following equation: J A 2ffn (10) where Afn is the initial offset frequency of the receiver.

Taking the example of a mobile handset, assuming the handset has a worst case frequency offset of Afn = +23hE],z, f5 = 270.833kHz, then equation 10 yields J c 5.88.

So J can be 2, 3, 4 or 5. In simulation 4 and 5 give better performance than 2 and 3.

The calculation of the arctangent of the AP can be carried out by any suitable known method. One suitable method is shown in EP-A- I 148642 and is known as CORDIC (coordinate rotation digital computation).

Claims

1. A method of estimating the phase offset between a frequency burst in a received digital signal and its expected phase, the method comprising sampling the signal in which the frequency burst is to be detected, detecting a frequency burst, calculating a centre complex correlation value for a number K of samples estimated to be at the centre of the frequency burst, the centre complex correlation value being calculated in accordance with the expression: n=K > Sbpf (n) Sbpf (n-J) n=0 wherein Sbpf (n) is the complex value of the nth sample which results from filtering the sample stream; sbPr is the complex conjugate of the filtered (n-J)th sample; and 2 5J, and deriving an accumulated phase error estimation (AP) from the centre complex correlation value.

2. A method as claimed in claim 1 in which 2 5 J 5 f' 2Afn in which is the sampling frequency, and In is a worst case frequency offset of the signal receiver.

3. A method as claimed in claim I in which the phase error estimate Alp is calculated from the equation arctangent (AP) = J 2 + J * Alp

f

4. A method as claimed in claim 3 in which the frequency offset Af between the received signa] and the transmitted signal is derived from the equation By = 2 * fs