JP3236226B2 - Echo canceller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller.

[0002]

2. Description of the Related Art In order to cancel a line echo generated in a hybrid circuit in a transmission device accommodating an international line, etc.
Generally, an echo canceller is used. The erasing characteristic of the echo canceller is deteriorated due to a change in the transfer characteristic of the echo path. For this reason, an adaptive filter having good convergence characteristics for generating a pseudo echo signal by following the fluctuation of the transfer characteristics of the echo path at high speed is required.

As a calculation algorithm for realizing such an adaptive filter having good convergence characteristics, there are the Kalman method and the RLS (recursive least squares) method shown in Document 1.

Literature 1, "Sakai," Recent Trends in Adaptive Algorithms-Focusing on RLS Method ", 1992, Journal of the Acoustical Society of Japan, Vol. 48, No. 7, pp. 493-500. End input signal x (n) and R
Tap coefficient h '(k) which is an estimated value of the transfer characteristic of the echo path obtained by the LS method (that is, h' (0) to h '(p)
) To generate a pseudo echo signal y (n) by the convolution operation shown in equation (1). In the equation (1), Σ represents the sum of k from 0 to p (p + 1 is the number of taps), and n represents the processing time.

Y (n) = {h ”(k) x (n−k) (1)

[0006]

As described above, in order to generate the pseudo echo signal y (n), the adaptive filter uses the tap coefficient h 'which is an estimated value of the transfer characteristic of the echo path.
(k) is required. Here, in the adaptive filter, the number of taps must be determined according to the transfer characteristics of the echo path (the characteristics of the impulse response).

However, since the transfer characteristics of the echo path to be calculated are unknown, the length of the impulse response of the echo path may not be specified in some cases.

In this case, a fixed number of taps estimated in advance with a margin to cover the impulse response of the echo path is used, but sometimes the number of taps longer than the necessary number of taps is used to generate the pseudo echo signal. Generation operation processing may be performed. Particularly, in the initial stage of the adaptive operation, the maximum calculation according to the fixed number of taps is performed even though the input signal is not sufficiently present, which is not efficient in terms of the amount of calculation.

It is known that the RLS method exhibits a fast convergence characteristic and converges by repetitive calculation within twice the number of taps. However, as the number of taps of the adaptive filter increases, the convergence becomes smaller. The number of repetitions for
As a result, a large amount of calculations must be performed.

Therefore, in the initial stage of the adaptive operation,
It is desired to increase the efficiency of the calculation for generating the pseudo echo signal.

[0011]

In order to solve such a problem, an echo canceller according to the present invention comprises (1) a near-end input signal on which an echo is superimposed, a far-end input signal corresponding to the number of taps, An adaptive filter unit that estimates the transfer characteristic of the echo path at that time from the internal state quantity and generates a pseudo echo signal by convolution of the estimated value and the far-end input signal, and (2) sequentially determines the transfer characteristic of the echo path. An internal state updating unit that updates an internal state quantity having a number of elements determined according to the number of taps necessary for estimation, and outputs the updated internal state quantity to the adaptive filter unit; (3) the echo canceller; It is determined whether or not to initialize, and in the case of initialization, the number of taps is set to an initial value, then the tap number is sequentially increased, and after the tap number reaches the preset number, the value is increased. Keep it constant A block size control unit that sequentially outputs the number of taps that change to the internal state update unit and the adaptive filter unit, and (4) an adder that subtracts the pseudo echo signal output from the adaptive filter unit from the near-end input signal. It is characterized by having.

With such a configuration, the amount of calculation at the beginning of the convergence operation when the adaptive filter section is initialized can be reduced, and high-speed convergence characteristics can be realized.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the echo canceller according to the present invention will be described below in detail with reference to the drawings.

This embodiment uses the RLS method as an algorithm for updating the estimated value of the transfer characteristic of the echo path. Therefore, hereinafter, a principle updating method of various values by the RLS method will be described.

Assuming that ER is a pseudo echo signal (echo replica) which is the output of the adaptive filter, the pseudo echo signal ER at time t is represented by a far-end input signal vector x _n (t) as shown in equation (2). , And a tap coefficient vector h ′ _n (t) which is an estimated value of the transfer characteristic of the echo path. Here, h ′ _n (t) is an n-dimensional vector (vertical vector) at time t, and x _n (t) is a vertical vector composed of sample values of far-end input signals from time t to the past n. Further, n is the number of taps of the adaptive filter unit, and this embodiment is characterized in that the number of taps n changes as described later.

[0016]

(Equation 1) In the RLS method, a tap coefficient vector h ′ _n (t), which is an estimated value of the transfer characteristic of the echo path at time t, is
Near-end input signal (scalar amount) y (t) at time t and n-dimensional gain vector (vertical vector) k _n (t) at time t
And the far-end input signal vector x _n (t) and the tap coefficient vector h ′ _n (t-1) at time t−1, which are updated by the recurrence formula shown in the formula (3). .

[0017]

(Equation 2) Here, the n-dimensional gain vector k _n (t) at time t is
As shown in equation (4), the n-dimensional internal state variable matrix P _n (t-1) at time t−1 and the far-end input signal vector x _n
(t) and a forgetting coefficient (scalar amount) λ for reducing the influence of past information.

[0018]

(Equation 3) The internal state variable matrix P _n (t) at time t is
As shown in equation (5), the far-end input signal x _n (t) and the time t
In the gain vector k _n (t), the time t-1 the state transition matrix in P _n (t-1), is intended to be updated sequentially by recurrence formula using a forgetting factor lambda.

[0019]

(Equation 4) As is clear from the above, in the RLS method, the time t
Is updated, the gain vector k is calculated according to equation (4).
_n (t) is updated, and the tap coefficient vector h ′ _n is updated using the updated gain vector k _n (t) according to the equation (3).
After updating (t), the pseudo echo signal ER at the time t is formed by using the updated tap coefficient vector h ′ _n (t) according to the equation (2). Further, the internal state variable matrix P _n (t) necessary for updating the gain vector k _n (t) at the next time is updated according to the equation (5).

Here, the various vectors and matrices used in the above equations (2) to (5) have dimensions determined by the number n of taps of the adaptive filter unit.

In the prior art, the number of taps n of the adaptive filter unit is fixed in any case. However, in this embodiment, the number of taps n Is updated gradually. Therefore, the dimensions of various vectors and matrices that are appropriately updated by the RLS method are also changed according to the change in the number n of taps.

In this embodiment, the number of taps is changed based on the following concept.

For example, when neither the far-end input signal nor the near-end input signal exists, the adaptive operation is restarted from the beginning by initializing the tap coefficient vector h ′ _n (t) of the adaptive filter section. In this case, when various vectors and matrices are calculated with a fixed number of taps n (here, n = p) as in the related art, even when the number of samples of the far-end input signal or the like is less than p, Performs operations on p-dimensional vectors and matrices. However, since the number of samples is less than p,
The number of elements that take valid values of vectors and matrices obtained as a result of the operation naturally decreases. In spite of a small number of elements having valid values, as a dimension corresponding to the number of taps,
It can be said that performing operations on vectors and matrices is wasteful. For example, in the internal state variable matrix P _n (t), there are elements of the square of the number of taps p, and an operation corresponding to this number of elements is necessary. There are few elements with large values, and many element operations are wasted.

Therefore, in this embodiment, when the number of taps used in the adaptive filter unit is to be redone from the beginning, the number of taps n is sequentially increased in accordance with the number of effective samples after initialization. We are trying to reduce.

FIG. 1 is a block diagram showing a functional configuration of the echo canceller according to the embodiment based on the above concept.

In FIG. 1, the far-end input signal Rin input to the far-end input terminal 10 is input to the echo canceller 1 of this embodiment, passes through the echo canceller 1, and is output from the output terminal 11 to the next stage. Provided to the processing circuit.
Far end input signal Rout that has passed through echo canceller 1
(Rin) leaks to the transmission line of the near-end input signal Sin input from the near-end input terminal 12 as an echo ET via an echo path (composed of a hybrid circuit or the like) 200 and becomes a near-end input signal Sin. Superimposed. The near-end input signal Sin ′ on which such an echo ET is superimposed is
The echo is canceled by being input to the echo canceller 1 of this embodiment, and the near-end input signal Sout (residual signal ZS) after the cancellation is output from the terminal 13.

The echo canceller 1 according to the embodiment having the input / output relationship as described above has the block size control unit 101
, Internal state updating section 102, adaptive filter section 103
And an adder R.

The adaptive filter section 103 includes a near-end input signal Sin ′ (= y (n)) at the current time t and a far-end input signal Rin.
A far-end input signal vector x _n (t) consisting of sample values for the number of taps p and a tap coefficient vector h ′ at the immediately preceding time
_Then, a tap coefficient vector h ′ _n (t) at the current time t is formed (estimating the transfer characteristic of the echo path) from _n (t−1) according to the above equation (3), and the tap coefficient vector h ′
_{From n} (t) and the far-end input signal vector x _n (t), (2)
The pseudo echo signal ER is generated according to the equation.

As described later, the adaptive filter section 103 is provided with a tap number control signal BC (= n) from the block size control section 101, and the tap number control signal BC (= n) is supplied to the adaptive filter section 103. A tap coefficient vector h ′ _n (t) having a corresponding dimension is generated. here,
When the dimension is to be increased, the following equation (8) is used.

The internal state updating unit 102 receives a far-end input signal Rin (a far-end input signal vector x _n (t) composed of n sample values corresponding to the number of taps). x _n (t), the internal state variable matrix P _n (t-1) and the forgetting factor (scalar amount) λ at the immediately preceding time t−1, which are internally held, and the current time t
Gain is intended to form the vector k _n (t), also the far-end input signal vector x _n (t), the updated gain vector k _n (t), the state transition in the immediately preceding time t-1 in The internal state variable matrix P _n (t) at the current time t is formed from the matrix P _n (t-1) and the forgetting factor λ according to the equation (5). The formed gain vector k _n (t) and internal state variable matrix P _n (t) at the current time t are internally stored in the internal state update unit 102, and the gain vector k _n (t) is This is provided to the filter unit 103.

As will be described later, the internal state updating unit 102 is also supplied with the tap number control signal BC (= n) from the block size control unit 101, and the tap number control signal BC (= n) Gain vector k of dimension according to
_n (t) and an internal state variable matrix P _n (t). Here, when the dimension of the internal state variable matrix P _n (t) is increased, the following equation (7) is used. Also,
The dimension of the gain vector k _n (t) is the internal state variable matrix P
_It increases when the dimension of _n (t) or the dimension of the far-end input signal vector x _n (t) increases.

The block size control section 101 determines whether or not the echo canceller 1 needs to be initialized based on the input state of the far-end input signal Rin and / or the near-end input signal Sin ′ (in other words, various types of signals). The timing of initializing the vectors and matrices h ′ _n (t), k _n (t), and P _n (t)), and if necessary, the tap number control that defines the number of taps After the signal BC (= n) is initialized to 1, the tap number control signal BC is incremented by one each time the time t is updated, and the tap number control signal BC reaches the preset maximum tap number p. Thereafter, the value of the tap number control signal BC is maintained at the maximum tap number p even when the time elapses. As described above, the tap number control signal B that is updated with the passage of time only immediately after initialization
C is an internal state update unit 102 and an adaptive filter unit 101
Output to

The adder R subtracts the pseudo echo signal ER from the adaptive filter unit 103 from the near-end input signal Sin 'on which the echo ET is superimposed, and cancels the echo.

The echo canceller 1 of this embodiment, which comprises the above components, operates as shown in the flowchart of FIG. 2 throughout. Note that the processing loop in FIG. 2 is performed once at each time.

When a new time t arrives, first, for example, the far-end input signal Rin is
And the near-end input signal Sin 'are monitored, and the operation mode is determined from the relationship between the two (step S1). For example, the pseudo echo signal ER is determined by whether the far-end input signal and the near-end input signal, or one of the two signals, is not input for a certain period of time (a state in which the audio level is not reached).
It is determined whether or not to perform the initialization of the adaptive operation for generating.

When performing initialization, the block number control unit 101 outputs a tap number control signal BC having a value of 1.
(= N) is the internal state update unit 102 and the adaptive filter unit 1
03, and at this time, the internal state updating unit 102
As shown in the equation (6), the internal state variable matrix P _n (t−1) at the immediately preceding time t−1 is set to a 1 × 1 matrix having only the initial parameter α (fixed value) (step S2).

P _n (t−1) = α (6) On the other hand, if it is determined in step S 1 that initialization is unnecessary, the block size control unit 101
Checks whether or not the tap number control signal BC (= n) up to now has already reached the maximum tap number p, and if not, it increments the tap number control signal BC (=
n) is replaced by the internal state update unit 102 and the adaptive filter unit 103
To perform the dimension expansion processing. On the other hand, when the maximum tap number p has been reached, the tap number control signal BC (= n) of the maximum tap number p is sent to the internal state updating unit 102 and the adaptive filter unit 103. The output is performed so that the dimension expansion processing is not performed (step S3).

Here, the internal state updating unit 10 to which the incremented tap number control signal BC (= n) is given.
2 is an internal state variable matrix P _n (t−1) at the immediately preceding time t−1
Is updated to a matrix whose dimension is increased by 1 according to the equation (7), and the tap number control signal B incremented by 1
The adaptive filter unit 103 given C (= n) converts the tap coefficient vector h ′ _n (t−1) at the immediately preceding time t−1 into (8)
According to the formula, the matrix is updated to a matrix whose dimension is larger by one.

[0039]

(Equation 5) Then, regardless of whether or not the dimension expansion operation is performed, the gain vector k _n (t) and the internal state variable at the current time t are calculated by the internal state updating unit 102 according to the above-described equations (4) and (5). A matrix P _n (t) is formed (step S4).
Subsequently, in the adaptive filter unit 103, the above (3)
According to the equation, the tap coefficient vector h ′ at the current time t
_n (t) is formed (step S5). It should be noted that the far-end input signal vector x _n (t) used in these arithmetic processings
Of the tap number control signal BC at that time.
(= N).

Then, in the adaptive filter unit 103,
The pseudo echo signal ER at the current time t is formed in accordance with the above-described equation (2), and the pseudo echo signal E
An echo canceling operation is performed by subtracting R (step S6).

When a series of processing at the current time t is completed in this way, the flow returns to step S1 to return to the next time (t = t
+1).

When initialization is required by the processing shown in FIG. 2 as described above, the adaptive operation with the effective tap number of 1 is performed (S1, S2, S4 to S6), and thereafter, the maximum tap number is set. An adaptive operation in which the number of effective taps is increased by one at each time until the number of taps reaches p (S1, S3 (with update of the number of taps), S4 to S6), and after reaching the maximum number p of taps, Perform an adaptive operation with a number p (S
1, S3 (without updating the number of taps), S4 to S6).

According to the above embodiment, the operation of the internal state updating unit and the adaptive filter unit is configured so that the number of taps is sequentially increased from the start of initialization in which the adaptive operation is restarted, so that it follows the transfer characteristic of the echo path. The amount of calculation at the time of initializing the adaptive operation can be reduced.

When the impulse response of the echo path is relatively short, the tap coefficients can be made to converge more quickly to the transfer characteristics of the echo path even if the amount of calculation is reduced in this way.

Further, various parameters such as tap coefficients of the adaptive filter section are inherited to the next time even when the number of taps is increased, so that the convergence operation is not interrupted.
A quick convergence operation can be realized.

In the above embodiment, the echo canceller applied to the RLS method has been described. However, the internal state quantity which is successively updated at each time represented by a vector or a matrix having a dimension corresponding to the number of taps is described. The present invention can be applied to any echo canceller that employs an algorithm for forming tap coefficients (vectors) using the algorithm.
For example, the present invention can be applied to an echo canceller using a Kalman filter method, a learning identification method, or the like.

Further, in the above embodiment, an echo canceller for canceling an echo due to impedance mismatch of the hybrid circuit is assumed, but the present invention is applied to an echo canceller for canceling an echo circulating from a speaker to a microphone. be able to.

[0048]

As described above, according to the present invention, the number of taps for the operation of the internal state update unit and the adaptive filter unit is sequentially increased from the start of initialization in which the adaptive operation is restarted. Can be reduced at the time of initialization of the adaptive operation to follow the transfer characteristic of the echo path, and the tap coefficient can be made to converge more quickly to the transfer characteristic of the echo path when the impulse response of the echo path is relatively short.

[Brief description of the drawings]

FIG. 1 is a functional block diagram illustrating a configuration of an embodiment.

FIG. 2 is a flowchart showing the operation of the embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Echo canceller, 101 ... Block size control part, 102 ... Internal state update part, 103 ... Adaptive filter part, R ... Adder.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04B 3/23 H03H 17/00 H03H 21/00

Claims (2)

(57) [Claims] 1. A transfer characteristic of an echo path at that time is estimated from a near-end input signal on which an echo is superimposed, a far-end input signal corresponding to the number of taps, and an internal state quantity. An adaptive filter unit that generates a pseudo echo signal by convolution operation with the end input signal, and updates the internal state quantity having the number of elements determined according to the number of taps necessary for sequentially estimating the transfer characteristic of the echo path Then, an internal state updating unit that outputs the updated internal state amount to the adaptive filter unit, and whether or not to initialize the echo canceller is determined.When the initialization is performed, the number of taps is set to an initial value. After that, the number of taps is sequentially increased, and after the number of taps reaches a preset number, the value is kept constant, and the number of taps changing in this way is determined by the internal state updating unit and the adaptive filter unit. And the block size control section that sequentially outputs, echo canceller, characterized in that it comprises an adder which subtracts the echo replica signal outputted from the adaptive filter unit from the near-end input signal. 2. The method according to claim 1, wherein the adaptive filter estimates a transfer characteristic of the echo path according to a sequential least squares method, and the internal state updating unit updates an internal state quantity according to a sequential least squares method. 2. The echo canceller according to 1.