Digital Signal Processing (DSP) is an emerging subject that involves many disciplines and is widely used in many fields. Since the 1960s, with the rapid development of computer and information technology, digital signal processing technology emerged and developed rapidly. In the past two decades, digital signal processing has been widely used in communications and other fields.
Digital signal processing is the use of computers or special processing equipment to collect, transform, filter, estimate, enhance, compress, and identify signals in digital form to obtain a signal form that meets people's needs. Digital signal processing is developed around the theory, implementation and application of digital signal processing. The theoretical development of digital signal processing has promoted the development of digital signal processing applications. Conversely, the application of digital signal processing has promoted the improvement of digital signal processing theory. The realization of digital signal processing is a bridge between theory and application
Digital signal processing is based on many disciplines, and its scope is extremely wide. For example, in the field of mathematics, calculus, probability and statistics, stochastic processes, and numerical analysis are all basic tools for digital signal processing, and are closely related to network theory, signal and system, cybernetics, communication theory, and fault diagnosis. Some newly emerging disciplines, such as artificial intelligence, pattern recognition, neural networks, etc., are inseparable from digital signal processing. It can be said that digital signal processing takes many classic theoretical systems as its theoretical basis, and at the same time makes itself the theoretical basis of a series of emerging disciplines.
The realization methods of digital signal processing are generally as follows:
(1) Realize with software (such as Fortran, C language) on a general-purpose computer (such as PC);
(2) Add a dedicated accelerated processor to the general-purpose computer system;
(3) It is realized by a general-purpose single-chip microcomputer (such as MCS-51, 96 series, etc.). This method can be used for some less complicated digital signal processing, such as digital control, etc.;
(4) Realize with a general programmable DSP chip. Compared with single-chip microcomputers, DSP chips have software and hardware resources more suitable for digital signal processing, and can be used for complex digital signal processing algorithms;
(5) Realize with a dedicated DSP chip. In some special occasions, the required signal processing speed is extremely high, which is difficult to achieve with general-purpose DSP chips, such as DSP chips dedicated to FFT, digital filtering, convolution, and related algorithms. This chip integrates the corresponding signal processing algorithms The chip is implemented in hardware without programming.
Among the above methods, the disadvantage of the first method is that it is slower and can generally be used for the simulation of DSP algorithms; the second and fifth methods are highly specific and their application is greatly restricted. The second method is also It is inconvenient for the independent operation of the system; the third method is only suitable for implementing simple DSP algorithms; only the fourth method opens up new possibilities for the application of digital signal processing
Although the theory of digital signal processing has developed rapidly, before the 1980s, due to the limitation of implementation methods, the theory of digital signal processing has not been widely used. It was not until the birth of the world's first single-chip programmable DSP chip in the late 1970s and early 1980s that the theoretical research results were widely applied to low-cost practical systems and promoted the development of new theories and application fields. It is no exaggeration to say that the birth and development of DSP chips have played a very important role in the technological development of communications, computers, control and other fields in the past 20 years.
In a DSP system, the input signal can have various forms. For example, it can be a voice signal output by a microphone or a modulated data signal from a telephone line, or a camera image signal that is encoded and transmitted on a digital link or stored in a computer.
The input signal is first subjected to band-limited filtering and sampling, and then A/D (Analog to Digital) conversion is performed to convert the signal into a digital bit stream. According to the Nyquist sampling theorem, in order to ensure that information is not lost, the sampling frequency must be at least twice the highest frequency of the input band-limited signal.
The input of the DSP chip is the digital signal expressed in sampling form obtained after A/D conversion. The DSP chip performs some form of processing on the input digital signal, such as a series of multiply and accumulate operations (MAC). Digital processing is the key to DSP, which is very different from other systems (such as telephone switching systems). In the switching system, the processor's role is to perform routing selection, and it does not modify the input data. Therefore, although both are real-time systems, their real-time constraints are quite different. Finally, the processed digital samples are converted into analog samples by D/A (Digital toAnalog) conversion, and then interpolation and smoothing filtering are performed to obtain continuous analog waveforms.
It must be pointed out that the DSP system model given above is a typical model, but not all DSP systems must have all the components in the model. For example, the voice recognition system is not a continuous waveform at the output, but the recognition result, such as numbers, text, etc.; some input signals are digital signals (such as CD Compact Disk), so there is no need to perform analog-to-digital conversion.
The digital signal processing system is based on digital signal processing, so it has all the advantages of digital processing:
(1) Convenient interface. DSP systems are compatible with other systems or devices based on modern digital technology. It is much easier to interface with such systems to implement certain functions than with analog systems to interface with these systems;
(2) Easy to program. The programmable DSP chip in the DSP system allows designers to modify and upgrade the software flexibly and conveniently during the development process;
(3) Good stability. The DSP system is based on digital processing, is less affected by ambient temperature and noise, and has high reliability;
(4) High precision. The 16-bit digital system can reach an accuracy of 10^(-5);
(5) Good repeatability. The performance of the analog system is greatly affected by the performance changes of the component parameters, while the digital system is basically not affected, so the digital system is convenient for testing, debugging and mass production;
(6) Convenient integration. The digital components in the DSP system are highly standardized, facilitating large-scale integration.
Of course, digital signal processing also has certain shortcomings. For example, for simple signal processing tasks, such as a telephone interface with an analog switch line, the use of DSP will increase the cost. The high-speed clock in the DSP system may cause problems such as high-frequency interference and electromagnetic leakage, and the DSP system consumes a lot of power. In addition, D SP technology is updated fast, requires many mathematical knowledge, and the development and debugging tools are not perfect.
Although the DSP system has some shortcomings, its outstanding advantages have made it more and more widely used in many fields such as communication, voice, image, radar, biomedicine, industrial control, and instrumentation.
In general, there is no very good formal design method for the design of DSP systems.
Before designing a DSP system, you must first determine the system's performance indicators and signal processing requirements according to the goals of the application system, which can usually be described by data flow diagrams, mathematical operation sequences, formal symbols or natural language.
The second step is to simulate the high-level language according to the requirements of the system. Generally speaking, in order to achieve the ultimate goal of the system, the input signal needs to be properly processed, and different processing methods will lead to different system performance. To get the best system performance, you must determine the best at this step. The processing method is the algorithm of digital signal processing (Algorithm), so this step is also called the algorithm simulation stage. For example, the speech compression coding algorithm is to obtain the best synthesized speech under a certain compression ratio. The input data used for algorithm simulation is obtained by collecting actual signals, and is usually stored as a data file in the form of a computer file. For example, the voice signal used in the simulation of the voice compression coding algorithm is actually collected and stored as a voice data file in the form of a computer file. The input data used in some algorithm simulations does not necessarily have to be the actual collected signal data. As long as the feasibility of the algorithm can be verified, it is also possible to input hypothetical data.
After completing the second step, the next step is to design a real-time DSP system. The design of a real-time DSP system includes hardware design and software design. The hardware design must first select the appropriate DSP chip according to the size of the system's calculations, the requirements for the calculation accuracy, the system cost restrictions, and the volume and power consumption requirements. Then design the peripheral circuit and other circuits of the DSP chip. Software design and programming are mainly based on the system requirements and the selected DSP chip to write the corresponding DSP assembly program. If the system has a small amount of calculation and is supported by a high-level language compiler, it can also be programmed in a high-level language (such as C language). Since the efficiency of existing high-level language compilers is not as efficient as that of manually writing assembly language, a mixed programming method of high-level language and assembly language is often used in actual application systems. The method of writing is to write assembly language, while the high-level language is used where the amount of calculation is not large. Using this method can not only shorten the software development cycle, improve the readability and portability of the program, but also meet the requirements of real-time operation of the system.
After the DSP hardware and software design is completed, it is necessary to debug the hardware and software. Debugging of software generally resorts to DSP development tools, such as software simulators, DSP development systems, or emulators. When debugging DSP algorithms, the method of comparing real-time results and simulation results is generally adopted. If the input of the real-time program and the simulation program are the same, the output of the two should be the same. Other software of the application system can be debugged according to the actual situation. Hardware debugging generally uses a hardware emulator for debugging. If there is no corresponding hardware emulator and the hardware system is not very complicated, it can also be debugged with the help of general tools.
After the software and hardware of the system are separately debugged, the software can be separated from the development system and run directly on the application system. Of course, the development of DSP system, especially software development, is a process that needs to be repeated. Although the performance of real-time system can be basically known through algorithm simulation, in fact, the simulation environment cannot be completely consistent with the real-time system environment. When migrating a simulation algorithm to a real-time system, it is necessary to consider whether the algorithm can run in real time. If the computational complexity of the algorithm is too large to run on the hardware in real time, the algorithm must be revised or simplified.
DSP chip, also known as digital signal processor, is a microprocessor especially suitable for digital signal processing operations. Its main application is to realize various digital signal processing algorithms in real time and quickly. According to the requirements of digital signal processing, DSP chips generally have the following main features:
(1) One multiplication and one addition can be completed in one instruction cycle;
(2) The program and data space are separated, and instructions and data can be accessed at the same time;
(3) There is fast RAM on-chip, which can usually be accessed simultaneously in two blocks through independent data buses;
(4) Hardware support with low overhead or no overhead loop and jump;
(5) Fast interrupt processing and hardware I/O support;
(6) Multiple hardware address generators that operate in a single cycle;
(7) Multiple operations can be performed in parallel;
(8) Support pipeline operation, so that operations such as fetching, decoding, and execution can be performed overlapped.
Of course, compared with general-purpose microprocessors, other general-purpose functions of DSP chips are relatively weak.
DSP chip development
The world's first single-chip DSP chip should be the S2811 released by AMI in 1978. The commercial programmable device 2920 released by Intel in 1979 was a major milestone for DSP chips. Neither chip has the single-cycle multiplier necessary for modern DSP chips. In 1980, the μP D7720 introduced by NEC Corporation of Japan was the first commercial DSP chip with a multiplier.
After that, the most successful DSP chips were a series of products from Texas Instruments (TI). TI successfully launched its first-generation DSP chip TMS32010 and its series products TMS32011, TMS320C10/C14/C15/C16/C17 in 1982, and then successively introduced the second-generation DSP chip TMS32020, TMS320C25/C26/C28, and the third Generation DSP chip TMS320C30/C31/C32, fourth generation DSP chip TMS320C40/C44, fifth generation DSP chip TMS320C5X/C54X, improved second generation DSP chip TMS320C2XX, high performance DSP chip TMS320C8X integrating multiple DSP chips And currently the fastest sixth-generation DSP chip TMS320C62X/C67X, etc. TI summarizes the commonly used DSP chips into three series, namely: TMS320C2000 series (including TMS320C2X/C2XX), TMS320C5000 series (including TMS320C5X/C54X/C55X), and TMS320C6000 series (TMS320C62X/C67X). Today, TI’s series of DSP products have become the most influential DSP chips in the world today. TI has also become the world's largest DSP chip supplier, and its DSP market share accounts for nearly 50% of the world's share.
The first to use CMOS technology to produce floating-point DSP chips was Japan's Hitachi company, which introduced floating-point DSP chips in 1982. In 1983, the MB8764 launched by Fujitsu, Japan, had an instruction cycle of 120ns and had dual internal buses, which made a big leap in processing throughput. The first high-performance floating-point DSP chip should be the DSP32 launched by AT&T in 1984.
Compared with other companies, Motorola is relatively late in launching DSP chips. In 1986, the company introduced the fixed-point processor MC56001. In 1990, it introduced the floating-point DSP chip MC96002 compatible with the IEEE floating-point format.
American Analog Devices (Analog Devices, AD for short) also occupies a certain share in the DSP chip market, and has successively introduced a series of DSP chips with its own characteristics. Its fixed-point DSP chips include ADSP2101/2103/2105, ASDP2111/2115, ADSP2161/2162/2164 and ADSP2171/2181, floating-point DSP chips include ADSP21000/21020, ADSP21060/21062, etc. Since 1980, DSP chips have been developed by leaps and bounds, and DSP chips have become more and more widely used. From the perspective of computing speed, the MAC (one multiplication and one addition) time has been reduced from 400ns (such as TMS32010) in the early 1980s to less than 10ns (such as TMS320C54X, TMS320C62X/67X, etc.), and the processing capacity has been increased by several times. The key multiplier components inside the DSP chip have dropped from about 40% of diearea in 1980 to less than 5%, and the amount of on-chip RAM has increased by more than an order of magnitude. In terms of manufacturing process, 4μm was adopted in 1980
The N-channel MOS (NMOS) process is generally adopted, but now the sub-micron (Micron) CMOS process is generally used. The number of pins of a DSP chip has increased from a maximum of 64 in 1980 to more than 200 now. The increase in the number of pins means an increase in structural flexibility, such as the expansion of external memory and communication between processors. In addition, the development of DSP chips has greatly reduced the cost, volume, weight, and power consumption of DSP systems. Table 1.1 is a comparison table of TI's DSP chips in 1982, 1992 and 1999. Table 1.2 is some data of representative chips from major DSP chip suppliers in the world.
DSP chips can be classified in the following three ways.
1. According to basic characteristics
This is classified according to the working clock and instruction type of the DSP chip. If at any clock frequency within a certain clock frequency range, the DSP chip can work normally, except for the change in the calculation speed, there is no performance degradation. This type of DSP chip is generally called a static DSP chip. For example, DSP chip of Japan OKI Electric Company, TMS320C2XX series chip of TI Company belong to this kind of category.
If there are two or more DSP chips, their instruction sets and corresponding machine code machine pin structures are compatible with each other, then this type of DSP chip is called a consistent DSP chip. For example, the TMS320C54X of TI of the United States falls into this category.
2. According to data format
This is classified according to the working data format of the DSP chip. DSP chips whose data work in a fixed-point format are called fixed-point DSP chips, such as TI’s TMS320C1X/C2X, TMS320C2XX/C5X, TMS320C54X/C62XX series, AD’s ADSP21XX series, AT&T’s DSP16/16A, and Motolora’s MC56000. Floating-point DSP chips that work in floating-point format are called floating-point DSP chips, such as TMS320C3X/C4X/C8X from TI, ADSP21XXX series from AD, DSP32/32C from AT&T, MC96002 from Motolora, etc.
The floating-point formats used by different floating-point DSP chips are not exactly the same. Some DSP chips use custom floating-point formats, such as TMS320C3X, while some DSP chips use IEEE standard floating-point formats, such as Motorola’s MC96002, FUJITSU’s MB86232 and ZORAN’s ZR35325, etc.
3. According to purpose
According to the purpose of DSP, it can be divided into general-purpose DSP chip and special-purpose DSP chip. General-purpose DSP chips are suitable for ordinary DSP applications. For example, a series of DSP chips of TI Company are general-purpose DSP chips. The dedicated DSP chip is designed for specific DSP operations, and is more suitable for special operations, such as digital filtering, convolution and FFT. For example, Motorola's DSP56200, Zoran's ZR34881, Inmos's IMSA100, etc. belong to the dedicated DSP chip .
This book mainly discusses general-purpose DSP chips.
The choice of DSP chip design DSP application system, choosing DSP chip is a very important link. Only when the DSP chip is selected, can the peripheral circuits and other circuits of the system be further designed. In general, the choice of DSP chip should be determined according to the actual application system needs. Different DSP application systems have different choices of DSP chips due to different application occasions and application purposes. Generally speaking, the following many factors should be considered when choosing a DSP chip.
1. The operating speed of the DSP chip.
Operation speed is one of the most important performance indicators of DSP chips, and it is also a major factor that needs to be considered when choosing DSP chips. The computing speed of DSP chips can be measured by the following performance indicators:
(1) Instruction cycle: the time required to execute an instruction, usually in ns (nanoseconds). For example, the instruction cycle of TMS320LC549-80 when the main frequency is 80MHz is 12.5ns;
(2) MAC time: the time of one multiplication plus one addition. Most DSP chips can complete a multiplication and addition operation in one instruction cycle. For example, the MAC time of TMS320LC549-80 is 12.5ns;
(3) FFT execution time: the time required to run an N-point FFT program. Since the operations involved in the FFT operation are very representative in digital signal processing, the FFT operation time is often used as an indicator to measure the computing power of the DSP chip;
(4) MIPS: That is, millions of instructions are executed per second. For example, the processing capacity of TMS320LC549-80 is 80 MIPS, that is, 80 million instructions can be executed per second;
(5) MOPS: That is, millions of operations are performed per second. For example, the computing power of TMS320C40 is 275 MOPS;
(6) MFLOPS: That is, millions of floating point operations are performed per second. For example, the processing capacity of TMS320C31 when the main frequency is 40MHz is 40 MFLOPS;
(7) BOPS: That is, one billion operations are performed per second. For example, the processing capacity of TMS320C80 is 2 BOPS.
2. The price of DSP chips.
The price of DSP chip is also an important factor to consider when choosing DSP chip. If an expensive DSP chip is used, even if the performance is high, its application range will definitely be limited, especially for civilian products. Therefore, according to the actual system application, an affordable DSP chip needs to be determined. Of course, due to the rapid development of DSP chips, the price of DSP chips tends to drop relatively quickly. Therefore, a slightly more expensive DSP chip is selected in the development stage. When the system is developed, its price may have dropped by half or more.
3. The hardware resources of the DSP chip.
The hardware resources provided by different DSP chips are different, such as the amount of on-chip RAM and ROM, externally expandable program and data space, bus interface, I/O interface, etc. Even if it is the same series of DSP chips (such as TI's TMS320C54X series), different DSP chips in the series have different internal hardware resources and can adapt to different needs.
4. The arithmetic accuracy of the DSP chip.
The word length of general fixed-point DSP chips is 16 bits, such as the TMS320 series. But some companies have 24-bit fixed-point chips, such as Motorola's MC56001. The word length of a floating-point chip is generally 32 bits, and the accumulator is 40 bits.
5. Development tools for DSP chips.
In the development process of DSP system, development tools are indispensable. Without the support of development tools, it is almost impossible to develop a complex DSP system. If there is the support of powerful development tools, such as C language support, the development time will be greatly shortened. Therefore, when choosing a DSP chip, attention must be paid to the support of its development tools, including software and hardware development tools.
6. The power consumption of the DSP chip.
In some DSP applications, power consumption is also a problem that requires special attention. For example, portable DSP devices, handheld devices, and DSP devices for field applications have special requirements for power consumption. At present, low-power, high-speed DSP chips powered by 3.3V have been widely used.
7. other.
In addition to the above factors, the choice of DSP chip should also consider the form of packaging, quality standards, availability, life cycle, etc. Some DSP chips may have multiple packaging forms such as DIP, PGA, PLCC, and PQFP. Some DSP systems may ultimately require industrial-grade or military-grade standards. When choosing, you need to pay attention to whether the selected chip has an industrial-grade or military-grade similar product. If the designed DSP system is not just an experimental system, but needs mass production and may have a life cycle of several years or even more than ten years, then you need to consider the supply of the selected DSP chip and whether it has the same or even longer Life cycle and so on.
Among the above-mentioned many factors, generally speaking, the price of fixed-point DSP chip is cheaper, the power consumption is lower, but the calculation accuracy is slightly lower. The advantages of floating-point DSP chips are high accuracy of operation and convenient programming and debugging in C language, but they are slightly more expensive and consume more power. For example, TI's TMS320C2XX/C54X series are fixed-point DSP chips, with low power consumption and low cost as its main features. The TMS320C3X/C4X/C67X is a floating-point DSP chip with high arithmetic accuracy, convenient programming in C language, and short development cycle, but at the same time its price and power consumption are relatively high.
The computational load of the DSP application system is the basis for determining the choice of a DSP chip with processing capacity. If the amount of calculation is small, you can choose a DSP chip with less processing power, which can reduce the system cost. On the contrary, a DSP system with a large amount of calculation must choose a DSP chip with strong processing capability. If the processing capability of the DSP chip cannot meet the system requirements, it must use multiple DSP chips for parallel processing. So how to determine the amount of calculation of the DSP system to select the DSP chip? Let's consider two cases below.
1. Sample processing
The so-called sample point processing is that the DSP algorithm loops once for each input sample point. This is the case with digital filtering. In digital filters, it is usually necessary to calculate once for each input sample point. For example, a 256-tap adaptive FIR filter using the LMS algorithm, assuming that the calculation of each tap requires 3 MAC cycles, the 256-tap calculation requires 256×3=768 MAC cycles. If the sampling frequency is 8kHz, that is, the interval between samples is 125ms, and the MAC cycle of the DSP chip is 200ns, 768 MAC cycles require 153.6ms, which obviously cannot be processed in real time, and a higher-speed DSP chip needs to be selected. Table 1.3 shows the processing requirements of the two signal bandwidths on the three DSP chips. The MAC cycles of the three DSP chips are 200ns, 50ns, and 25ns, respectively. It can be seen from the table that the latter two DSP chips can be implemented in real time for the application of the dialogue belt. For audio applications, only the third DSP chip can process in real time. Of course, in this example, no other calculations are considered.
2. Processing by frame Some digital signal processing algorithms do not loop once for each input sample, but loop once every certain time interval (usually called a frame). For example, the medium and low speed speech coding algorithm usually takes 10ms or 20ms as a frame, and the speech coding algorithm loops once every 10ms or 20ms. Therefore, when choosing a DSP chip, you should compare the processing capacity of the DSP chip in a frame with the calculation amount of the DSP algorithm. Suppose the instruction cycle of the DSP chip is p (ns), and the time of one frame is Dt
(Ns), then the maximum amount of calculation that the DSP chip can provide in one frame is Dt/p instructions. For example, the instruction cycle of TMS320LC549-80 is 12.5ns, and if the frame length is 20ms, the maximum amount of operations that TMS320LC549-80 can provide in one frame is 1.6 million instructions. Therefore, as long as the calculation amount of the speech coding algorithm does not exceed 1.6 million instructions, it can be run in real time on the TMS320LC549-80.
Application of DSP chip
Since the birth of DSP chips in the late 1970s and early 1980s, DSP chips have developed rapidly. The rapid development of DSP chips has benefited from the development of integrated circuit technology on the one hand and the huge market on the other. In the past 20 years, DSP chips have been widely used in many fields such as signal processing, communications, and radar. At present, the price of DSP chips is getting lower and lower, and the performance-price ratio is increasing day by day, which has huge application potential. The main applications of DSP chips are:
(1) Signal processing-such as digital filtering, adaptive filtering, fast Fourier transform, correlation calculation, spectrum analysis, convolution, pattern matching, windowing, waveform generation, etc.;
(2) Communication-such as modem, adaptive equalization, data encryption, data compression, echo cancellation, multiplexing, fax, spread spectrum communication, error correction coding, video phone, etc.;
(3) Voice-such as voice coding, voice synthesis, voice recognition, voice enhancement, speaker identification, speaker confirmation, voice mail, voice storage, etc.;
(4) Graphics/images-such as two-dimensional and three-dimensional graphics processing, image compression and transmission, image enhancement, animation, robot vision, etc.;
(5) Military-such as confidential communications, radar processing, sonar processing, navigation, missile guidance, etc.;
(6) Instruments and meters-such as spectrum analysis, function generation, phase-locked loop, seismic processing, etc.;
(7) Automatic control-such as engine control, voice control, automatic driving, robot control, disk control, etc.;
(8) Medical treatment-such as hearing aids, ultrasound equipment, diagnostic tools, patient monitoring, etc.;
(9) Household appliances-such as high-fidelity audio, music synthesis, tone control, toys and games, digital phones/TVs, etc.
With the continuous improvement of the performance-price ratio of DSP chips, it is foreseeable that DSP chips will be more widely used in more fields.
Our other product:
