1. The development history of DSP
The foundation of informatization is digitization. One of the core technologies of digitization is digital signal processing. The task of digital signal processing needs to be completed by DSP devices to a large extent. DSP technology has become a cutting-edge technology that people are paying more and more attention to and getting rapid development. DSP can stand for Digital Signal Processor (Digital Signal Processor) or Digital Signal Processing (Digital Signal Processing). The latter is a theoretical technology, and the former must be transformed into an actual product. The combination of the two becomes a means to solve a certain practical problem and realize a certain program, namely, the digital signal processing solution (DSPS). This article focuses on the first interpretation of DSP-digital signal processor. DSP is a dedicated processor that performs high-speed real-time processing after the analog signal is converted into a digital signal. Its processing speed is 10-50 times faster than the fastest CPU. In today's digital age, DSP has become a basic device in fields such as communications, computers, and consumer electronics. Industry insiders predict that DSP will be the fastest-growing electronic product in integrated circuits in the future, and it will become a decisive factor in the upgrading of electronic products.
Before the advent of DSP, digital signal processing could only be done by MPU (microprocessor). But the lower processing speed of MPU cannot meet the requirements of high-speed real-time. Therefore, in the 1970s, someone proposed the theory and algorithm basis of DSP. And DSP only stays in the textbook, even the developed DSP system is composed of discrete components, and its application field is limited to the military and aviation sectors.
With the development of large-scale integrated circuit technology, the world's first DSP chip was born in 1982. This DSP device is made of micron process NMOS technology. Although the power consumption and size are slightly larger, the calculation speed is dozens of times faster than MPU, and it is widely used in speech synthesis and codec. The advent of DSP chips marks a big step forward for DSP application systems from large-scale systems to miniaturization. With the progress and development of CMOS technology, the second generation of DSP chips based on CMOS technology came into being, and its storage capacity and computing speed were doubled, becoming the foundation of voice processing and image hardware processing technology. In the late 1980s, the third-generation DSP chip came out, and the computing speed was further improved, and its application range was gradually expanded to the fields of communications and computers.
DSP developed the fastest in the 1990s, and the fourth and fifth generations of DSP devices appeared one after another. The current DSP belongs to the fifth generation of products. Compared with the fourth generation, it has a higher system integration level, integrating the DSP core and peripheral components on a single chip. This highly integrated DSP chip not only shows its talents in the fields of communications and computers, but also gradually penetrates into the field of people's daily consumption. The prospects are very promising.
2. Features and advantages of DSP
1. Hardware features
(1) DSP belongs to Modified Harvard architecture, that is, it has two internal buses: data bus and program bus. The program and the data storage space are separated, each has an independent address bus and data bus, and the fetching and reading can be performed at the same time. At present, it has reached 9 billion floating-point operations per second (9000MFLOPS).
(2) Use streamlined operations. The execution of each instruction is divided into several steps such as instruction fetching, decoding, fetching, and execution, which are respectively completed by multiple functional units on the chip. It is equivalent to the parallel execution of multiple instructions, which greatly improves the computing speed.
(3) Independent hardware multiplier. Multiplication instructions are completed in a single cycle, optimizing a large number of repeated multiplications in algorithms such as convolution, digital filtering, FFT, correlation, and matrix operations.
(4) Circular addressing (Circular addressing), bit-reversed (bit-reversed) and other special instructions greatly improve the addressing, sorting and calculation speed in operations such as FFT and convolution. The 1024-point FFT time is already less than 1μs.
(5) Independent DMA bus and controller. There are one or more groups of independent DMA buses, which work in parallel with the CPU's program and data bus. The DMA speed has reached over 800Mbyte/s without affecting the CPU's work.
(6) Multi-processor interface. It is convenient for multiple processors to work in parallel or serial to increase the processing speed.
(7) JTAG (Joint Test Action Group) standard test interface (IEEE 1149 standard interface). It is convenient for on-chip online simulation of DSP and debugging under multi-DSP conditions.
2. Software features
(1) Immediate addressing: The operand is an immediate value, which can be obtained directly from the instruction. Example: MOV A, @0x16; send the constant 0x16 to register A.
(2) Direct addressing: For example, TI’s TMS320 series chip divides the data memory into 512 pages, each with 128 words. Set a data page pointer DP (Data Pointer), use 9-bit to point to a data page, and add a 7-bit offset address in the page to form a 16-bit data address. This helps to speed up the addressing speed.
(3) Indirect addressing: ① 8 auxiliary registers, one auxiliary register pointer designates an auxiliary register arithmetic unit for 16-bit unsigned number operation, determines a new address, and loads one of the auxiliary registers. ②The contents of the 8 auxiliary registers are quite flexible, and they can be loaded, added, or subtracted from immediate data; they can be loaded from the data memory; they can also be used for some indexed addressing. ③Because of the reverse position, the bit reverse addressing can be realized.
(4) Unique multiplication instruction: Example: MAC X0, Y0, AX: (R0) +, X0 Y: (R4) + N4, YO This instruction commands DSP56300: Multiply the numbers in registers X0 and Y0, and add the result In Acc A, load the value of the memory address pointed to by register R0 into register X0, load the value of Y memory address pointed to by register R4 into register Y0, and add 1 to the value of R0, and add the value of register N4 to R4.
The advantages of DSP system based on digital signal processing compared with traditional analog signal processing system:
(1) The interface is simple and convenient. Due to the simple electrical characteristics of the digital signal, it is easy to implement on the hardware interface when different DSP systems are connected to each other. On the data stream interface, each system only needs to follow a specific standard protocol.
(2) High precision and good stability. The digital signal processing is only affected by the dualization error and the limited word length, and the processing process does not introduce other noises, so it has a higher signal-to-noise ratio. In addition, the performance of the analog system is greatly affected by the performance of the component parameters, while the digital system is basically unchanged, so the digital system is more convenient for testing, debugging and mass production.
(3) Convenient programming and easy implementation of complex algorithms. In the DSP system, the DSP chip provides a high-speed computing platform, and the system functions depend on software programming. When it is combined with modern signal processing theory and computational mathematics, it can realize complex digital signal processing functions.
(4) Convenient integration. Modern DSP chips integrate the DSP core and its peripheral circuits on a single chip. This structure facilitates the design of portable and highly integrated digital products.
In addition, modern DSP chips, as programmable very large integrated circuits (VLSI) devices, implement digital signal processing functions through downloadable software and firmware. In addition to the calculation and control functions of ordinary microprocessors, DSP chips have also made great improvements in the processor structure, instruction system and instruction flow design for high data transmission rate and real-time digital signal processing with intensive numerical calculations.
3. Problems and challenges faced by DSP
The maturing DSP still has many areas for improvement, but also faces many challenges.
(1) How to rationally arrange the data flow so that it can be executed smoothly without conflict among the execution units of the DSP is still an important problem faced by DSP developers. Due to the complexity of the design, when mapping the algorithm to the DSP specific target hardware, high-level programming languages cannot be used yet, assembly language must be used, and a very clear understanding of the parallel execution mechanism of the device. And this programming design limited to assembly language is the bottleneck to improve the efficiency of software development.
(2) There are still problems with parallel structures. In order to achieve higher throughput, more data bits must be processed in a specific unit of time. VLIW technology represents the parallelism of the instruction level. The superscalar structure and the superpipeline structure also try to get more instructions in one instruction cycle. Data level parallelism is represented by wider data words, vectorization and data flow structures. Because the width of the data word is larger, each instruction cycle instruction can process more data, which increases the number of data bits that can be processed in each clock cycle. Task-level or transaction-level parallelism is reflected in multitasking, multithreading, and multiprocessor designs. These structures are expected to increase data processing throughput, but the increased data and instruction width and the consequent increase in data processing throughput will have to pay a certain price. When the code density and data width match the application, they can help, but when the data word width is not the same as the processor, they can cause a lot of trouble instead.
(3) A large number of available on-chip caches are becoming more and more important to the total throughput of the system, because standard memory buses and interfaces can no longer provide support for the gigabyte data transfer rate of each MAC in the system. Whether the rest of the system can be compatible with high-speed processors is also becoming a big problem. A dual MAC processor with 2 ALU units may require 4 data words per clock cycle, or more than 4 gigabytes of data per second. word.
(4) The challenges faced by the development of DSP are also reflected in the rapid increase in CPU speed and the continuous decline in prices. DSP manufacturers are faced with two choices, one is to accelerate the development of DSP, and the other is to withdraw from the competition. Each manufacturer must shift from diversified investment to unitary investment, and establish products with DSPS as the main development, that is, integrating all technologies and all products in DSP.
4. the development trend and prospects of DSP
DSP continues to meet people's ever-increasing requirements on its development road, and is gradually developing in the direction of personalization and low power consumption. The prospects for DSP development are very impressive.
(1) System-level integration of DSP is the trend
Reducing the size of DSP chips is always the direction of DSP technology development. Most of the current DSPs are based on a RISC (reduced instruction set computing) structure. The advantages of this structure are small size, low power consumption, and high performance. Various DSP manufacturers have adopted new processes to improve DSP cores, and integrate several DSP cores, MPU cores, special processing units, peripheral circuit units, and storage units on a single chip to become DSP system-level integrated circuits.
(2) Programmable DSP is the leading product
Programmable DSP provides manufacturers with great flexibility. Manufacturers can develop various series of products with different models on the same DSP platform to meet the needs of different users. At the same time, programmable DSP also provides a good way for users to easily upgrade.
(3) Fixed-point DSP is the mainstream
In theory, although floating-point DSP has a larger dynamic range than fixed-point DSP and is more suitable for DSP applications, fixed-point arithmetic DSP devices have lower cost, lower memory requirements, and lower power consumption. . Therefore, programmable DSP devices for fixed-point arithmetic are still mainstream products on the market. According to statistics, most of the DSP devices currently sold are 16-bit fixed-point programmable DSP devices, and it is expected that the proportion will gradually increase in the future.
(4) Pursue higher computing speed
At present, the general DSP operation speed is 100MIPS, that is, it can operate 100 million instructions per second. Due to the personalization and customization trend of electronic equipment, DSP must pursue higher and faster computing speeds in order to keep up with the pace of updating electronic equipment. The increase of DSP operation speed mainly depends on the new technology to improve the chip structure. Currently, DSP devices mostly use 0.5μm～0.35μm CMOS technology. According to the development trend of CMOS, it is completely possible that the operation speed of DSP can be increased by 100 times (to 1600GIPS).
5. concluding remarks
The increasing application of DSP in various fields has driven the development of DSP itself. Its application in 3C (Communications, Computers, Consumer) fields has occupied 90% of the current DSP market share, indicating that DSP has huge potential in other fields. . In the future development, it will appear in various fields with better performance.
