Design Feature: August 1, 1996

Echoes during a conversation are the bane of mobile-cellular, cordless, and speaker phones. Echoes are unacceptable in a quality-speech telephone, and you must eliminate them to ensure normal conversation. The causes of echoes include: acoustic feedback from speaker to microphone in a hands-free system; impedance mismatches between a two-wire local loop and four-wire telephone network, causing electrical signals to reflect; and constant, prolonged background noise present in moving vehicles or at public spaces, such as emergency telephones along highways.

Figure 1 shows the common location of the echo-canceller function in a telephone handset. Traditionally, the echo-cancellation function required software development on general-purpose DSP µPs. However, you can now use dedicated echo-canceller chips to eliminate echoes from these sources. Figure 2, which you can use in either a handset or base set, eliminates software development. In addition, the design can also reduce power dissipation through lower system-clock speeds. This design provides a 30-dB cancellation level for echoes of as high as 50 msec introduced by acoustic and line feedback.

The following equation mathematically expresses the echo signal, Y_n, where T is the sample cycle, X_n is the receiving signal at T_n , and W_i is the impulse response of the echo path:

The definition of the pseudo echo signal [mean]Y_n is

and the residual echo signal E_n is

where H_i are the tap coefficients of the echo-canceller chip. The sampling, arithmetic, and digital synthesis operations are all implemented on chip. For example, the chip automatically updates the tap coefficients to reduce the error between the received echo signal and the pseudo echo signal generated on chip.

In Figure 2's circuit, two sources induce echoes: acoustic feedback from R_OUT to S_IN on the acoustic (left-hand) side, and the line feedback from S_OUT to R_IN at the line (right-hand) side. Echo-canceller IC₂ provides acoustic cancellation, and echo-canceller IC₃ provides line cancellation. If an application needs only acoustic or line-echo cancellation, only one of these ICs is necessary. Other than their input and output connections, the two echo cancellers are identical and can eliminate echoes over a 27-msec time period. To cancel echoes longer than 27 msec, you can use a related IC that provides echo cancellation over 55-msec time periods. You can also cascade multiple devices to cancel echoes with delays in excess of 55 msec. The design also uses PCM codecs IC1 and IC4 to perform bidirectional analog/digital conversion.

The MSM7602 echo-canceller IC contains a digital transverse adaptive filter that estimates the impulse response by observing the transfer characteristics of the echo path. About 300 samples are necessary to fully characterize the impulse response of the echo path. To cancel received echoes at the chip, this chip generates a pseudo echo by calculating the tap coefficients of the filter, which cancels the actual received echo.

On the acoustic side of Figure 2, S_IN and R_OUT connect to a microphone and speaker, respectively. The CLK input to the circuit is the external input for the basic clock. RST and PWDWN are both active-low signals for reset and power-down support. The External SCK input is the clock-signal input that controls the transmitting and receiving of serial data at rates of 64 to 2048 kHz. The External SYNC signal provides an external input for synchronizing transmit and receive functions at a typical rate of 8 kHz.

The values of R₃ and R₂—according to the equation 1+R₂/R₃—produce the desired gain for the IC₂'s input stage. A gain of less than 10 is typical to implement a stable circuit. R₁ and C₁ determine the low-frequency response of the audio signal. This network passes frequencies of 30 Hz or greater (f=1/2[pi]R₁C₁). R₄ pulls up the open-drain MOS output device on PCMOUT. A value of 2.2 k(ohm) can sink 0.5 mA and provide a fast response characteristic. C₂ and C₃ are low- and high-frequency bypass capacitors, respectively. R₅ and R₆ are pull-up resistors that force logical one when the output stage of IC₂ is inactive.

Similarly, on the line side, R₈ and R₉ determine the gain for IC₃'s codec. C₅ and R₇ determine this codec's low-frequency response. R₁₀ is the pull-up resistor for the open-drain PCMOUT output, and C₆ and C₇ provide low- and high-frequency bypass, respectively. Pull-up resistors R₁₁ and R₁₂ force logical ones when the output stage of IC3 is inactive.

Tests results verify operation

Figure 3 presents the results from a typical test of the echo-cancellation circuit, using a 2-kHz, 2V p-p sine wave at R_IN on the line side. This test was the result of directly connecting the speaker output at the acoustic side to the microphone input to simulate 100% feedback at the acoustic side of the handset. Figure 3 shows the output waveform, on the line side at S_OUT, with and without the echo cancellation. The test verifies an echo cancellation of the signal of 30 dB. You can observe similar results if you provide feedback at the line side by tying R_IN to S_OUT, using S_IN as the input signal, and observing R_OUT at the acoustic side.

Jamal Sarma is a senior applications engineer in the Logic Products Group at Oki Semiconductor, Sunnyvale, CA. He works on a range of telecommunications products including echo cancellers, codecs, wireless LANs, and PC cards. He holds a BSEE from Osmania University, India, an MSEE degree from Stanford University (Stanford, CA) and an MBA from Pepperdine University (Los Angeles). You can reach Sarma at sarma@okisemi.com.

References

1. Noguchi, Osamu and Takizawa, Yumi, "Hands-free service for cellular system," Oki Technical Review, #127, July 1987, vol 53, pg 26.