Anti-interference technology in speech signal processing system based on DSP

1. Introduction

At present, DSP has become a basic device in the fields of communications, computers, and consumer electronics due to its fast computing speed, rich on-chip resources, and the ability to implement complex linear and nonlinear algorithms. It is particularly prominent in speech signal processing technology. However, since all electronic devices, including the DSP itself, are interference sources, and there are many external interference sources in the working environment of the system, and speech recognition technology is very sensitive to signal noise, the anti-interference problem of the system must be considered in system design. Otherwise, it will at least affect the processing results of the system, and even cause more serious consequences. This article introduces the anti-interference technology in the speech signal processing system based on DSP.

2 System interference sources and interference paths

The interference sources in the DSP-based voice signal processing system mainly include atmospheric noise caused by lightning discharge, atmospheric noise caused by sunspot movement, thermal noise caused by heat generated by electronic components such as resistors, grid interference caused by 50 Hz power frequency power grid, interference caused by household appliances, brush interference caused by motors, ignition system interference caused by automobile ignition devices, radio frequency interference caused by wireless communication systems, and interference caused by some malicious human factors. Among all the interference sources, high-frequency pulse noise is the most harmful to digital signal processing systems (audio interference caused by voice is not studied here).

The interference sources mentioned above are all electromagnetic interference (EMI). Electromagnetic principles: As long as there is current, a magnetic field will be generated, and as long as there is voltage, an electric field will be generated. The amount of change in the magnetic field and electric field over time is the root cause of electromagnetic interference. The concept of electromagnetic interference is shown in Figure 1.

The interference paths of the DSP-based voice signal processing system mainly include power lines, input/output lines, ground lines, electromagnetic induction, electrostatic induction, common impedance of the circuit, power supply anomalies, etc. The proportion of various interference paths in the system is shown in Table 1.

3 Anti-interference measures

Based on the analysis of the system itself, interference sources and interference paths, the main anti-interference measures are:

① Improve the electromagnetic compatibility of the system itself;

②Isolate the interference source;

③Cut off interference paths.

Based on these three solutions, this paper proposes some hardware anti-interference technologies and software anti-interference technologies suitable for this system.

3.1 Electromagnetic compatibility

Electromagnetic compatibility means that when power, electronic, communication equipment or its system is in working condition at the place where it is installed, it will not affect its surroundings, nor be affected by the electromagnetic environment around it, will not cause malfunction and performance degradation, and obtain its working capacity as designed. In other words, the equipment or system does not generate electromagnetic interference from the outside world, and its normal working ability is not affected by the electromagnetic interference in the environment.

3.2 Hardware Anti-interference Technology

Since high-frequency pulse noise is the most harmful to this system, in order to improve the anti-interference performance of the system, the following measures can be taken in the system:

(1) Increase the anti-interference ability of the bus. Use a three-state gate bus structure and connect a pull-up resistor to the bus to keep the bus at a stable high level in an instant and avoid the bus being suspended. A buffer must be added to the bus.

(2) Improve the anti-interference ability of the system control signal. There are usually control lines such as RESET and STB in the system. When the transmission distance between the CPU and its control device is far and the control line impedance is high, it is easy to be interfered by pulse noise. A 20 pF capacitor can be connected in parallel to the input end of the controlled device, and a 0.01 μF capacitor can be connected in parallel to the control signal line such as RESET. Adding a buffer to the control line can reduce the impedance of the control line and also help suppress interference.

(3) Suppress the crosstalk interference of digital signals. This crosstalk interference is caused by adjacent signal lines during signal transmission, and mostly occurs on parallel conductors on printed circuit boards. The strength of crosstalk interference is related to the coupling impedance between two adjacent signal lines and the impedance of the signal itself. Therefore, in this system, the length of the signal line should be shortened as much as possible; when transmitting multiple level signals, the level signals with similar leading and trailing edge times should be grouped together for transmission as much as possible; a larger ground area should be arranged on the back of the double-sided printed circuit board to absorb and shield the high-frequency pulse noise generated by the components.

(4) Use electromagnetic interference filters (EMI Filters) to eliminate power supply interference. With the emergence and widespread popularity of electronic devices, computers, and household appliances, power grid interference is becoming increasingly serious and has become a public nuisance. In particular, transient electromagnetic interference has a high voltage amplitude (several hundred volts to thousands of volts), a fast rise rate, a short duration, and strong randomness, which can easily cause serious interference to digital circuits and even damage equipment. Electromagnetic interference filters, also known as power supply noise filters (PNFs), can effectively suppress power grid noise, improve the anti-interference ability of equipment and the reliability of the system. The application of electromagnetic interference filters in the system is shown in Figure 2.

(5) Use the hardware watchdog function to improve the system's anti-interference ability. The watchdog circuit composed of the dedicated device MAX692 is shown in Figure 3. The system uses few peripheral components. MAX692 is a microsystem monitoring circuit with functions such as backup battery switching, power-off detection, and watchdog monitoring. Among them, WDI is the watchdog detection input terminal, which is connected to a dedicated I/O port or a bus port of the DSP. RESET is the reset signal output terminal, which is connected to the reset terminal RST of the DSP. The WDI timing period of MAX692 is 1.6 s, and the reset pulse width is 200 ms. If WDI remains high or low for more than the "watchdog" timing period (1.6 s), the RESET terminal will generate a negative pulse with a width of 200 ms (minimum 140 ms) to reset the DSP.

3.3 Software Anti-interference Technology

Software can also be used to improve the anti-interference ability of the DSP voice processing system. The main ones are:

① Use digital filters to filter out interference;

② Use software watchdog, multiple sampling technology, timed output refresh and other technologies to suppress interference.

The following mainly introduces the application of digital filter in this system.

The processing process of the digital filter (DF) on the speech signal is shown in Figure 4. The speech signal first passes through the sample/hold circuit (S/H), is sent to the analog/digital converter (ADC) to be converted into a digital quantity, and then the interference signal is filtered out by the digital filter, and finally the speech signal output is obtained by the digital/analog converter (DAC).

According to the different mathematical models used, digital filters can be divided into two categories: one is a recursive filter, whose characteristic is that the output of the filter is not only related to the input signal, but also to the past output value; the other is a non-recursive filter (such as first-order and second-order low-pass filters), whose characteristic is that the output of the filter is only related to the input signal, but not to the past output value. This system uses a recursive filter.

Assume that the input signal of the digital filter is X(n) and the output signal is Y(n), then the relationship between the input sequence and the output sequence can be expressed by the difference equation:

In the above formula, the input signal X(n) can be a digital sequence obtained after sampling and ADC conversion of the speech signal, or it can be the output signal of the computer; ak and bk are coefficients. By setting ak and bk, DF can be designed into the required bandpass filter.

There are six methods for software design of digital filters: program judgment filtering method (limited filtering method), median filtering method, arithmetic average filtering method, recursive average filtering method, pulse interference prevention average filtering method, and first-order filtering method. According to the needs, this system selects the program judgment filtering method, and the design process is shown in Figure 5.

4 Conclusion

Experiments have shown that the above anti-interference methods can fully suppress interference from outside the system in the DSP-based speech signal processing system, and effectively improve the system's anti-interference ability and reliability.