KR0134268B1 - Improved least mean square method - Google Patents

Abstract

The present invention relates to an improved LMS (LMS) method, by placing a plurality of convergence conditions during the execution of the SMS, it is determined that the convergence if any one is satisfied by terminating the LS, accurately determining whether the convergence of the SMS It is possible to speed up the execution of LS.

Description

Improved LMS Method

1 is a schematic block diagram of a conventional ghost removal device.

FIG. 2 is a detailed block diagram of the filter unit and filter coefficient coefficient unit of FIG.

3 is a flow chart illustrating the flow of an improved LS method according to the present invention.

* Explanation of symbols for main parts of the drawings

10: AD converter 20: reference signal extractor

30: filter coefficient calculation unit 40: filter unit

41: delay unit 42: FIR filter

43: IIR filter 50: DA converter

The present invention relates to an LS method, and more particularly, to an improved LS method capable of adaptively responding to an SMS convergence condition.

In general, a video signal from a broadcasting station is reflected by an obstacle such as a mountain or a building while being received by a receiving antenna, or reflected without an impedance matching of a transmission path, thereby forming a multipath channel (MULTIPLE CHANNEL). For this reason, the reflected video signal in addition to the original video signal appears on the monitor of the television is called Ghost.

FIG. 1 is a schematic block diagram of a conventional ghost elimination device, which includes an AD converter 10, a reference signal extractor 20, a filter coefficient calculator 30, a filter 40, and a DA converter 50. It is composed of

In the figure, the AD converter 10 converts an input video signal into a digital signal, and the reference signal extractor 20 extracts a ghost removal reference signal inserted into a predetermined portion of the digital signal converted into a digital signal. . Then, the filter coefficient calculation unit 30 calculates and then transmits the filter coefficients using the ghost elimination reference signal stored in advance and the ghost elimination reference signal extracted by the reference signal extracting unit 20, and the DA converter 50 The filtered video signal is converted into an analog signal again.

2 is a detailed block diagram of the filter coefficient calculation unit 30 and the filter unit 40. As can be seen from the figure, the filter unit 40 includes a delay unit 41 for delaying the original video signal to remove the line ghost, a FIR filter 42 for removing the near ghost and the line ghost, and a long ghost. It is composed of an IIR filter 43 and filters the input video signal using the filter coefficient transmitted from the filter coefficient calculation unit 30 and outputs it to the DA converter 50.

Hereinafter, the principle of removing the ghost generated in the transmission path will be described in more detail with reference to FIGS. 1 and 2.

When a broadcast station inserts and sends a ghost removal reference signal to a predetermined portion of the video signal, the video signal input from the AD converter 10 in the ghost removal apparatus is converted into a digital signal and output to the filter unit 40, and the reference signal extraction unit By 20, the ghost removal reference signal inserted in the predetermined portion of the video signal is extracted.

The filter coefficient calculation unit 30 compares the ghost removal reference signal stored in advance with the ghost removal reference signal extracted by the reference signal extraction unit 20 to calculate a filter coefficient that can remove the ghost generated in the transmission path. 40 is transmitted to the FIR filter 42 and the IIR filter 43, and the input image signals are filtered by the two filters 41 and 42 using the transmitted filter coefficients, thereby generating ghosts generated in the transmission path. Removed.

On the other hand, there are two methods of obtaining a filter coefficient having an inverse characteristic of channel characteristics: a sequential correction method and a daylight correction method.

First, the batch correction method has a short convergence time and no problem of convergence, while the calculation amount is large and the error in the steady state is relatively large. For example, in the case of using FFT (FAST FOURIER TRANSFORM), which is one of the batch correction methods, it depends on the DSP (DIGITAL SIGNAL PROCESSOR) used, but it corresponds to the channel characteristics within 1 second of data input time and 0.1-0.3 seconds. It is possible to calculate the filter coefficient for ghost removal and send it to the filter section.

Secondly, the sequential correction method is small in computation and simple in formula, easy to implement, and small in steady-state error, but has a convergence problem and takes a lot of time. For example, when using the LMS (LEAST MEAN SQUARE) method, which is one of the sequential correction methods, the filter coefficient is calculated for 1 second data input time, 2-4 seconds for short, and 10 seconds for long. Time is required.

If the LMS method is expressed by a formula, it is as follows.

From here,

e [n] = d [n] -y [n]

a _k [n + 1] = a _k [n] +2 μe [n] X [n] k], 0 ≦ k ≦ N1-1

b _k [n + 1] = b _k [n] +2 μe [n] y [nk], 1 ≦ _k ≦ N2

In Equation (1), x [n] is a ghost removal reference signal input to the ghost elimination apparatus through the transmission path, and d [n] is an original ghost elimination reference signal previously stored in the ghost elimination apparatus. . And a0... aN-1 is a filter coefficient of the FIR filter whose number of filter coefficients is N1, and b1... bN2 is the filter coefficient of the IIR filter whose filter coefficient is N2. In addition, PG is the number of filter coefficients for removing pre-ghost from the FIR filter, and μ is a gain constant (GAIN CONSTANT) related to convergence stability and speed.

Typically, the period of one line in the video signal is 1 / 15.734264KMz-63.55556μS, and the sampling period of the AD converter 10 is four times the frequency of the subcarrier Fxc = 3.579545 MHz, and 1 / (3.579545E6 × 4) = Sampling one line at 0.06984127871 μS yields 910 samples of discretization data. The number of samples of the discretized ghost removal reference signal is 336 samples.

The conventional LMS method performs the LMS for a large number of times to sufficiently converge, or terminates the LMS when e [n] in Equation 1 is smaller than the predetermined convergence value given one convergence condition. How to use.

However, ignoring the ghost characteristics caused by various channel characteristics and repeating the same number of times in a batch or determining convergence with only one convergence condition is not only time-consuming, but it is also possible to obtain an incorrect filter coefficient. There are a lot.

For example, the LMS performs LMS on all data from n = 0 to 909, as shown in Equation 1 above, but it takes a long time to repeat the LMS uniformly for a sufficient number of times until convergence. There is this.

In addition, the method of terminating the LMS when one is satisfied using one convergence condition is improved in time delay compared to repeating the LMS uniformly as described above. If the filter coefficient is not obtained, and the value is smaller than the predetermined convergence value, the final error is larger than the reference value even though the convergence value is already converged.

In addition, the magnitude of the final error varies depending on the channel characteristics, and it is inefficient to determine whether the final error is converged as one reference value.

Accordingly, an object of the present invention is to provide an improved LMS method for accurately determining convergence by applying various LMS convergence conditions according to channel characteristics. have.

In order to achieve the above object, the present invention, in the LMS method for calculating the filter coefficient for ghost removal by repeatedly updating the filter coefficient according to a predetermined convergence condition, the reference signal to be converged, the current filtered output signal and An improved LMS method is characterized by setting a plurality of convergence conditions by using a previous filtered output signal, and terminating the performance of the LMS when any one of the plurality of convergence conditions is satisfied. .

Other objects and various advantages of the present invention will become more apparent from the preferred embodiments of the invention described below with reference to the accompanying drawings by those skilled in the art.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

3 is a flowchart showing the operation of the LMS corresponding to the rapidly changing ghost according to the present invention, which will be described in more detail with reference to the accompanying drawings.

First, a convergence condition to be converged is set (step 310). Converging conditions according to a preferred embodiment of the present invention are as follows.

In equations (2), (3) and (4), y [n] is an output signal obtained as a result of performing LMS, and r [n] is a ghost removal reference signal to be converged at present. And, e1 _cur [n] is the current e1 [n], e1 _pre [n] is the difference between the output signal obtained as a result of the previous LMS and the ghost removal reference signal to converge. In addition, (4) is for measuring the temporal change rate of (2).

Next, the LMS is performed in a state where a plurality of convergence conditions are set (step 320), and a first difference in which the difference between the ghost elimination reference signal and the filter to be converged and the output signal is set during the LMS is performed. It is determined whether or not it is smaller than a predetermined value (step 330), and when it is satisfied, the LMS is terminated.

If the expression (2) is not satisfied in step 330, it is determined whether the condition of the expression (3), that is, the difference between e1 _pre [n] and e1 _cur [n] is smaller than the set second predetermined value (step 340) If satisfied, the LMS ends. At this time, (Equation 3) is to measure the relative change size rather than the absolute magnitude of the error, to correspond to the ghost shape having a relatively large magnitude of the error after convergence.

If the condition of Equation 3 is not satisfied in step 340, it is determined whether the condition of Equation 4 can be converged in a short time by measuring the temporal rate of change of e1 [n] (step 350). If satisfied, the LMS ends. If not satisfied, the LMS is repeated (step 320).

Accordingly, as described above, the present invention can not only accurately determine whether or not convergence is provided by placing several LMS convergence conditions, but also can quickly perform an LMS execution speed.

On the other hand, in the embodiment of the present invention has been described by determining whether or not the convergence by setting the three convergence conditions as described above, it will be apparent to those skilled in the art that the present invention is not limited thereto. .

Therefore, the present invention may further improve the accuracy of convergence when further adding conditions other than the above-mentioned convergence conditions while considering an appropriate convergence processing speed.

As described above, by using the present invention, various convergence conditions according to channel characteristics can be applied, whereby it is possible to accurately determine whether the LMS converges as compared to the conventional method described above, and also has an effect of speeding up the LMS performance. have.

Claims (2)

A LMS method for calculating filter coefficients for ghost elimination by repeatedly updating a filter coefficient according to a predetermined convergence condition, comprising: using a plurality of reference signals, a current filtered output signal, and a previous filtered output signal. Improved convergence (LMS) method, characterized in that it establishes a convergence condition and terminates the LMS execution when any one of the plurality of convergence conditions is satisfied. The method of claim 1, wherein the plurality of convergence conditions comprise: a first convergence condition that determines whether a first difference between a reference signal to be converged previously stored and an output signal passing through the filter is smaller than a first predetermined value; A second convergence condition for determining whether a difference between an output signal passing through a filter immediately before the current signal and a reference signal is smaller than a second predetermined value set by comparing the first difference if the first convergence condition is not satisfied; And a third convergence condition for determining a rate of change of the first convergence condition if the second convergence condition is not satisfied.