Let you name the name of the chip you know. You may not be able to remember for a moment or list all the DSP chips. Perhaps it is the deep understanding of the DSP chip that is clearly understood, forgotten due to various reasons; perhaps because everyone is saying that the DSP chip is good, since everyone says it is good, it is really good, as to how good it may be Understand. Well, whether you understand it or not, now let us read the world of this chip from a new perspective and let you discover the details that we did not understand.
DSP chip, also called digital signal processor, adopts a special hardware and software structure. It is a microprocessor focused on digital signal processing and operation. Its main application is realizing various digital signal processing in real time and is digital signal processing. An important technical tool for the theoretical practical application process. In the voice processing, image processing and other technical fields have been widely used. What is the most important difference from other chips based on the understanding of the DSP chip? Huang Tian, ​​an advanced embedded development manager at Hangzhou Hikvision Digital Technology Co., Ltd., believes that the biggest difference between DSP chips and other chips is that it has a large number of dedicated instructions designed for various algorithms, such as various vector operations. In addition, DSP chips take more consideration of the bandwidth and throughput of the data bus when designing, and avoiding data access becomes a bottleneck that affects the performance of the algorithm.
The basic structure of the chip
In order to quickly implement digital signal processing operations, DSP chips generally use special hardware and software structures. The following briefly introduces the basic structure of the DSP chip.
(1) The main characteristic of Harvard architecture is that programs and data are stored in different storage spaces. That is, program memory and data memory are two independent memories. Each memory is independently addressed and accessed independently. Corresponding to the two memories, the program bus and data bus are set in the system, so that the data throughput is doubled. Since the program and data are in two separate spaces, the fetch and execution can completely overlap.
(2) The pipeline operation pipeline is related to the Harvard architecture. The DSP chip widely uses the pipeline to reduce the instruction execution time, thereby enhancing the processing capability of the processor. The processor can process two to four instructions in parallel, each at a different stage in the pipeline. Listed below is an example of a three-stage pipeline operation:
CLLOUT1
Take the finger NN-1N-2;
Decoding N-1NN-2;
Perform N-2N-1N,
(3) A dedicated hardware multiplier for a dedicated hardware multiplier. The faster the multiplication speed, the higher the DSP processor performance. With a dedicated application multiplier, multiplication can be completed in one instruction cycle.
(4) Special DSP instructions DSP uses special instructions.
(5) Fast instruction cycle Special DSP instructions, DSP chips use special instructions. Fast instruction cycles, Harvard architecture, pipeline operations, dedicated hardware multipliers, and special DSP instructions, coupled with the optimized design of the integrated circuit, allow the DSP chip's instruction cycle to be less than 200 ns.
DSP System Features
Digital signal processing is different from common scientific calculations and analysis. It emphasizes the real-time nature of arithmetic processing. Therefore, in addition to the high-speed arithmetic and control functions emphasized by ordinary microprocessors, DSPs deal with real-time digital signal processing in the processor architecture. The instruction system and the instruction flow have many new features. Their characteristics are as follows:
The arithmetic unit has a hardware multiplier and a multi-function arithmetic unit. The hardware multiplier can perform the multiplication operation in a single instruction cycle, which is an important sign of the difference between the DSP and the general-purpose microprocessor. The multi-function operation unit can perform operations such as addition and subtraction, logic, shift, and data transfer. The new generation of DSP even contains multiple parallel computing units to increase its processing power. For filtering, correlation, matrix operations and other features that require a large number of multiply and accumulate operations, the multipliers and adders of the arithmetic unit of the DSP can perform two operations of multiplying and accumulating in one clock cycle. In recent years, some DSPs such as ADSP2106X and DSP96000 series DSPs can perform multiplication, addition, and subtraction at the same time, greatly speeding up the FFT butterfly speed.
Bus architecture Traditional general-purpose processors use a unified program and data space, shared program and data bus architecture, the so-called Von Neumann architecture. DSP generally adopts the Harvard architecture separated from the data bus and the program bus or the improved Harvard architecture, which greatly improves the instruction execution speed. Multiple sets of busses on the chip can simultaneously perform instruction fetching and multiple data access operations. Many DSP chips have built-in DMA controllers to match the on-chip multi-bus architecture and greatly increase the data block transfer speed.
The dedicated addressing unit DSP is intended for data-intensive applications. With frequent data access, the calculation of the data address also takes a lot of time. The DSP is internally equipped with a dedicated addressing unit for address modification and updating. They can automatically modify the content before or after addressing the access to point to the next address to be accessed. The modification and updating of the address work in parallel with the arithmetic unit and does not require extra time. The DSP address generator supports direct and indirect addressing operations. Most DSPs also support bit-reversed addressing (for FFT algorithms) and circular addressing (for digital filtering algorithms).
On-chip memory for the needs of data-intensive digital signal processing operations, DSP on the program and data access time is very demanding, in order to reduce the transfer time of instructions and data, many DSPs integrate high-speed program memory and data memory to improve Program and data access memory speed.
Most of the pipeline technology DSPs use pipeline technology, which means that the execution of one instruction is decomposed into several stages such as instruction fetching, decoding, number fetching, and execution. Each stage is called primary flow. Each instruction fetches, fetches, fetches, and executes operations by multiple functional units on the chip, thereby reducing the execution time of each instruction without increasing the clock frequency.
Difference between DSP and other processors
The difference between digital signal processors (DSPs), general-purpose microprocessors (MPUs), and microcontrollers (MCUs) is that DSPs are designed for high-performance, repetitive, and numerically computationally intensive real-time processing; MPUs are heavily used in computers; The MCU is suitable for control-based processing.
Benefits of DSP Chips The DSP's computing speed is much higher than other processors. Take FFT as an example. The high-performance DSP not only handles 4 to 10 times the processing speed of the MPU, but also can continuously complete the data real-time input/output. The DSP architecture is relatively single, and it is generally programmed in assembly language. The predictability of the task completion time is much more complex than the structure and instruction (super-scalar instructions) and MPU that are heavily dependent on the compilation system. Taking a FIR filter as an example, for each input of data, the filter coefficient corresponding to each stage needs one multiplication, one addition, one fetch, and two fetches, and a special data movement operation is needed. The DSP can perform multiplication in a single cycle. Plus parallel operations and 3 to 4 data access operations, and ordinary MPU to complete the same operation requires at least 4 instruction cycles. Therefore, under the same instruction cycle and on-chip instruction cache, the DSP operation speed can exceed 4 times the MPU operation speed.
DSP chip floating point and fixed point
When choosing a DSP device, does it use floating point or fixed point, if the fixed point is 16 or 32 bits? In fact, this problem is related to the dynamic range of the signal required by your algorithm.
The dynamic range of the floating-point DSP compared to the fixed-point DSP (dynamic range: the logarithm of the ratio between the maximum undistorted output power when the sound system is playing back and the noise output power of the system at rest, and the maximum output of a multimedia hard disk player. The relative ratio between light and darkest parts is much larger. Each 1 bit increase in the word length of a fixed-point DSP increases the dynamic range by 6dB, and the dynamic range of a 16-bit word length is 96dB. Programmers must always pay attention to the occurrence of overflow. For example: When doing image processing, the image is rotated, moved, etc., it is easy to produce overflow. At this time, either constantly shift calibration or truncate. The former consumes a lot of program space and execution time, and the latter quickly leads to deterioration of image quality.
In short, it is the performance of the entire system. Similar problems occur when dealing with low SNR signals, such as speech recognition, radar, and sonar signal processing. The dynamic range of the 32-bit floating-point operation DSP can reach 1536dB, which not only greatly expands the dynamic range, improves the operation accuracy, but also greatly saves the operation time and storage space, because the calibration, shift and overflow check are greatly reduced.
Since floating-point DSP floating-point operations are implemented in hardware and can be completed in a single cycle, the processing speed is much higher than that of fixed-point DSPs. This advantage is especially prominent when implementing high-precision complex algorithms.
The fixed-point calculation just treats a data as an integer. Usually AD samples are integers. This number has a scale factor relative to the real analog signal. Everyone knows to use a 16-bit AD to sample a 0 to For a 5V signal, the AD output divided by 2^16 and then 5V is the corresponding voltage. In a fixed-point DSP, this 16-bit sample is processed directly, and it is not converted to a voltage expressed in decimals, because a fixed-point DSP cannot represent a decimal with sufficient precision, and it can only calculate integers.
The advantage of a floating-point DSP is that it can convert this sampled integer to a fractional voltage without loss of precision (the decimal is represented in scientific notation) because the scientific notation can represent large dynamics. A signal of range, taking IEEE754 floating point number as an example, single-precision floating-point format: [31] 1-bit symbol [30-23] 8-bit index [22-00] 23 decimal places. The smallest number that can be represented is +-2^-149, the largest number is +-(2-2^23)*2^127, and the dynamic range is 20*log (the largest number/the smallest number). Such a large dynamic range of 1667.6dB makes it almost impossible to consider multiplication and accumulated overflow during programming, and if a fixed-point processor is used to program, rounding and shifting the calculation result is a common practice, which will lose precision to some extent. .
The reason is that the fixed-point processed signal has a limited dynamic range. For example, a 16-bit fixed-point DSP can represent integers ranging from 1-65536 and its dynamic range is 20*log(65536/1)=96dB. For a 32-point DSP, the dynamic range is 20 *log(2^32/1)=192dB, which is much less than the 1667.6dB of the 32-bit ieee floating-point number, but actually 192dB is sufficient for most applications. Due to the limited number of digits of the AD converter, the dynamic range of the input signal is generally small, but in the signal processing of the DSP, since the dot product operation increases the dynamic range of the intermediate node signal, the intermediate result in the signal processing flow is mainly considered. The dynamic range, as well as the algorithm's accuracy requirements for the intermediate results, select the appropriate DSP. In addition, the floating-point DSP is easier to program. In the fixed-point DSP programming, the programmer must constantly adjust the PQ value of the intermediate result. In fact, the intermediate result is constantly adjusted and rounded.
Real-value operations can be added directly to hardware operations through code, and fixed-point components must be software-indirect to perform real-number operations. This increases algorithm instructions and extends development time.
On the whole, fixed-point DSPs have advantages in cost and floating-point DSPs are better in ease of use.
Since the birth of DSP chips, DSP chips have been rapidly developed. On the one hand, thanks to the development of integrated circuits, on the other hand, it benefits from the huge market. In just over a decade, DSP chips have been widely used in signal processing, communications, and many other fields.
For DSP chip development status and how DSP and other products with application solutions Hikvision Huang Tian made the following observations: DSP chip has been in the direction of specialization, diversification, the market division of various manufacturers more and more Fine, the difference is also growing. In addition, simple DSP chips have been rare, and more is that DSP chips are integrated with other processing cores to form a highly integrated and targeted SOC, which not only greatly reduces board space, but also brings about The overall advantages of power consumption, cost, and development cycle have driven the development of the industry and improved product performance.
The advantage of DSP lies in flexible algorithm integration, which can provide products with powerful performance and flexible customization. The same product can implement different solutions for different customer needs. In order to improve the competitiveness of the products, manufacturers will make full use of algorithms in algorithms, algorithms will become more and more complex, but the stability of the algorithm, the power consumption of the products, and the development cycle will become uncontrollable risks. The DSP algorithm is not a pile of theoretical formulas, but is a sophisticated software that closely integrates with the specific features of the DSP chip used.
Development Status and Application of DSP Chips
These factors need to be fully considered when designing products. Do not blindly adopt so-called advanced algorithms and high-performance DSPs for some gimmick functions. Instead, start with user requirements and find the best combination of algorithms and DSPs. In the product solution, the algorithm and the DSP are the core. This combination determines that, with other processing chips and peripherals, an efficient product solution can be formed.
At present, the use of DSP technology from the military to civilian use, from aerospace to production and life, are increasingly using DSP. DSP technology is mainly used for radar and sonar signal processing in aerospace; in communications, it is mainly used for signal transmission of mobile phones, IP phones (Voiceover IP), ADSL, and HFC; in terms of control, it is mainly used for motor control, Optical drives and hard disk drives; for test/measurement, mainly used for virtual instruments, automatic test systems, medical diagnostics, etc.; in terms of electronic entertainment, mainly for high-definition television (HDTV), set-top box (STB), AC-3, Home theater, DVD and other applications;
In image/graphics, it is mainly used for two-dimensional and three-dimensional graphics processing, image compression and transmission, image enhancement, animation, etc.; as well as digital cameras, web cameras, and so on, all apply DSP technology. At the same time, SOC chip systems, wireless applications, and embedded DSPs are all directions and trends of DSP development in the future. It can be said that without DSP, there will be no access to the Internet, no multimedia, and no wireless communication. Therefore, DSP will still be the technological driver of the entire semiconductor industry. Nowadays, DSP applications continue to expand, and its coverage includes broadband Internet access services, the development of next-generation wireless communication systems, digital consumer electronics markets, and automotive electronics market development.
DSP chip classification
DSP chips specifically designed for different algorithms can be divided into three basic categories: basic characteristics, data formats, and applications.
According to the basic characteristics, according to the DSP chip operating clock and the type of instruction to be classified. If the DSP chip can work normally at any frequency within a certain clock frequency range, there is no performance degradation except for the change in calculation speed. Such a DSP chip is generally called a static DSP chip. If there are two or more kinds of DSP chips, their instruction set and the corresponding machine code machine pin structure are compatible with each other, this type of DSP chip is called a consistent DSP chip.
According to the data format, this is based on the data format of the DSP chip. The DSP chip whose data operates in a fixed-point format is called a fixed-point DSP chip. The work in floating-point format is called a DSP chip. The floating-point format used by different floating-point DSP chips is not exactly the same, some DSP chips use a custom floating-point format, and some DSP chips use the IEEE standard floating-point format.
According to the purpose, it can be divided into general-purpose DSP chips and special-purpose DSP chips. The general-purpose DSP chip is suitable for ordinary DSP applications, such as TI's series of 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.
Conclusion
With the rapid development of the security industry, especially the blowout of high-definition and intelligent demand, the entire industry has fully entered the digital era. As the core of digital security products, security chips have become a huge industry, so more different algorithms are needed. . For the ever-increasing algorithm requirements, the processing power of the DSP is never enough. Algorithm developers need to constantly balance the performance metrics and the processing power of the DSP. It is to make one cycle less, or to allow occasional frame loss, similar to this. If you can't make a choice, you need to further optimize the algorithm, either on the algorithm's architecture or at the assembly code level. On-chip cache is an important factor that affects the performance of the algorithm. DDR bandwidth is often system-level.
Due to the DSP's ability to calculate and not be good at logic processing, especially network protocols and database management, DSPs are often used in conjunction with other chips. Among the various development projects in the field of security, DSP+ARM is the most popular. ARM is responsible for network, storage, and peripheral management. DSP is responsible for image and audio processing and encoding and decoding. At present, DSP+ARM is basically integrated into an SOC. However, for software development, DSP and ARM are also separate. In areas where higher processing performance is required, multi-chip DSPs are often used together for processing. These are also Hikvision Huang Tian, ​​a profound experience of many years of industry senior technical people.
DSP chip, also called digital signal processor, adopts a special hardware and software structure. It is a microprocessor focused on digital signal processing and operation. Its main application is realizing various digital signal processing in real time and is digital signal processing. An important technical tool for the theoretical practical application process. In the voice processing, image processing and other technical fields have been widely used. What is the most important difference from other chips based on the understanding of the DSP chip? Huang Tian, ​​an advanced embedded development manager at Hangzhou Hikvision Digital Technology Co., Ltd., believes that the biggest difference between DSP chips and other chips is that it has a large number of dedicated instructions designed for various algorithms, such as various vector operations. In addition, DSP chips take more consideration of the bandwidth and throughput of the data bus when designing, and avoiding data access becomes a bottleneck that affects the performance of the algorithm.
The basic structure of the chip
In order to quickly implement digital signal processing operations, DSP chips generally use special hardware and software structures. The following briefly introduces the basic structure of the DSP chip.
(1) The main characteristic of Harvard architecture is that programs and data are stored in different storage spaces. That is, program memory and data memory are two independent memories. Each memory is independently addressed and accessed independently. Corresponding to the two memories, the program bus and data bus are set in the system, so that the data throughput is doubled. Since the program and data are in two separate spaces, the fetch and execution can completely overlap.
(2) The pipeline operation pipeline is related to the Harvard architecture. The DSP chip widely uses the pipeline to reduce the instruction execution time, thereby enhancing the processing capability of the processor. The processor can process two to four instructions in parallel, each at a different stage in the pipeline. Listed below is an example of a three-stage pipeline operation:
CLLOUT1
Take the finger NN-1N-2;
Decoding N-1NN-2;
Perform N-2N-1N,
(3) A dedicated hardware multiplier for a dedicated hardware multiplier. The faster the multiplication speed, the higher the DSP processor performance. With a dedicated application multiplier, multiplication can be completed in one instruction cycle.
(4) Special DSP instructions DSP uses special instructions.
(5) Fast instruction cycle Special DSP instructions, DSP chips use special instructions. Fast instruction cycles, Harvard architecture, pipeline operations, dedicated hardware multipliers, and special DSP instructions, coupled with the optimized design of the integrated circuit, allow the DSP chip's instruction cycle to be less than 200 ns.
DSP System Features
Digital signal processing is different from common scientific calculations and analysis. It emphasizes the real-time nature of arithmetic processing. Therefore, in addition to the high-speed arithmetic and control functions emphasized by ordinary microprocessors, DSPs deal with real-time digital signal processing in the processor architecture. The instruction system and the instruction flow have many new features. Their characteristics are as follows:
The arithmetic unit has a hardware multiplier and a multi-function arithmetic unit. The hardware multiplier can perform the multiplication operation in a single instruction cycle, which is an important sign of the difference between the DSP and the general-purpose microprocessor. The multi-function operation unit can perform operations such as addition and subtraction, logic, shift, and data transfer. The new generation of DSP even contains multiple parallel computing units to increase its processing power. For filtering, correlation, matrix operations and other features that require a large number of multiply and accumulate operations, the multipliers and adders of the arithmetic unit of the DSP can perform two operations of multiplying and accumulating in one clock cycle. In recent years, some DSPs such as ADSP2106X and DSP96000 series DSPs can perform multiplication, addition, and subtraction at the same time, greatly speeding up the FFT butterfly speed.
Bus architecture Traditional general-purpose processors use a unified program and data space, shared program and data bus architecture, the so-called Von Neumann architecture. DSP generally adopts the Harvard architecture separated from the data bus and the program bus or the improved Harvard architecture, which greatly improves the instruction execution speed. Multiple sets of busses on the chip can simultaneously perform instruction fetching and multiple data access operations. Many DSP chips have built-in DMA controllers to match the on-chip multi-bus architecture and greatly increase the data block transfer speed.
The dedicated addressing unit DSP is intended for data-intensive applications. With frequent data access, the calculation of the data address also takes a lot of time. The DSP is internally equipped with a dedicated addressing unit for address modification and updating. They can automatically modify the content before or after addressing the access to point to the next address to be accessed. The modification and updating of the address work in parallel with the arithmetic unit and does not require extra time. The DSP address generator supports direct and indirect addressing operations. Most DSPs also support bit-reversed addressing (for FFT algorithms) and circular addressing (for digital filtering algorithms).
On-chip memory for the needs of data-intensive digital signal processing operations, DSP on the program and data access time is very demanding, in order to reduce the transfer time of instructions and data, many DSPs integrate high-speed program memory and data memory to improve Program and data access memory speed.
Most of the pipeline technology DSPs use pipeline technology, which means that the execution of one instruction is decomposed into several stages such as instruction fetching, decoding, number fetching, and execution. Each stage is called primary flow. Each instruction fetches, fetches, fetches, and executes operations by multiple functional units on the chip, thereby reducing the execution time of each instruction without increasing the clock frequency.
Difference between DSP and other processors
The difference between digital signal processors (DSPs), general-purpose microprocessors (MPUs), and microcontrollers (MCUs) is that DSPs are designed for high-performance, repetitive, and numerically computationally intensive real-time processing; MPUs are heavily used in computers; The MCU is suitable for control-based processing.
Benefits of DSP Chips The DSP's computing speed is much higher than other processors. Take FFT as an example. The high-performance DSP not only handles 4 to 10 times the processing speed of the MPU, but also can continuously complete the data real-time input/output. The DSP architecture is relatively single, and it is generally programmed in assembly language. The predictability of the task completion time is much more complex than the structure and instruction (super-scalar instructions) and MPU that are heavily dependent on the compilation system. Taking a FIR filter as an example, for each input of data, the filter coefficient corresponding to each stage needs one multiplication, one addition, one fetch, and two fetches, and a special data movement operation is needed. The DSP can perform multiplication in a single cycle. Plus parallel operations and 3 to 4 data access operations, and ordinary MPU to complete the same operation requires at least 4 instruction cycles. Therefore, under the same instruction cycle and on-chip instruction cache, the DSP operation speed can exceed 4 times the MPU operation speed.
DSP chip floating point and fixed point
When choosing a DSP device, does it use floating point or fixed point, if the fixed point is 16 or 32 bits? In fact, this problem is related to the dynamic range of the signal required by your algorithm.
The dynamic range of the floating-point DSP compared to the fixed-point DSP (dynamic range: the logarithm of the ratio between the maximum undistorted output power when the sound system is playing back and the noise output power of the system at rest, and the maximum output of a multimedia hard disk player. The relative ratio between light and darkest parts is much larger. Each 1 bit increase in the word length of a fixed-point DSP increases the dynamic range by 6dB, and the dynamic range of a 16-bit word length is 96dB. Programmers must always pay attention to the occurrence of overflow. For example: When doing image processing, the image is rotated, moved, etc., it is easy to produce overflow. At this time, either constantly shift calibration or truncate. The former consumes a lot of program space and execution time, and the latter quickly leads to deterioration of image quality.
In short, it is the performance of the entire system. Similar problems occur when dealing with low SNR signals, such as speech recognition, radar, and sonar signal processing. The dynamic range of the 32-bit floating-point operation DSP can reach 1536dB, which not only greatly expands the dynamic range, improves the operation accuracy, but also greatly saves the operation time and storage space, because the calibration, shift and overflow check are greatly reduced.
Since floating-point DSP floating-point operations are implemented in hardware and can be completed in a single cycle, the processing speed is much higher than that of fixed-point DSPs. This advantage is especially prominent when implementing high-precision complex algorithms.
The fixed-point calculation just treats a data as an integer. Usually AD samples are integers. This number has a scale factor relative to the real analog signal. Everyone knows to use a 16-bit AD to sample a 0 to For a 5V signal, the AD output divided by 2^16 and then 5V is the corresponding voltage. In a fixed-point DSP, this 16-bit sample is processed directly, and it is not converted to a voltage expressed in decimals, because a fixed-point DSP cannot represent a decimal with sufficient precision, and it can only calculate integers.
The advantage of a floating-point DSP is that it can convert this sampled integer to a fractional voltage without loss of precision (the decimal is represented in scientific notation) because the scientific notation can represent large dynamics. A signal of range, taking IEEE754 floating point number as an example, single-precision floating-point format: [31] 1-bit symbol [30-23] 8-bit index [22-00] 23 decimal places. The smallest number that can be represented is +-2^-149, the largest number is +-(2-2^23)*2^127, and the dynamic range is 20*log (the largest number/the smallest number). Such a large dynamic range of 1667.6dB makes it almost impossible to consider multiplication and accumulated overflow during programming, and if a fixed-point processor is used to program, rounding and shifting the calculation result is a common practice, which will lose precision to some extent. .
The reason is that the fixed-point processed signal has a limited dynamic range. For example, a 16-bit fixed-point DSP can represent integers ranging from 1-65536 and its dynamic range is 20*log(65536/1)=96dB. For a 32-point DSP, the dynamic range is 20 *log(2^32/1)=192dB, which is much less than the 1667.6dB of the 32-bit ieee floating-point number, but actually 192dB is sufficient for most applications. Due to the limited number of digits of the AD converter, the dynamic range of the input signal is generally small, but in the signal processing of the DSP, since the dot product operation increases the dynamic range of the intermediate node signal, the intermediate result in the signal processing flow is mainly considered. The dynamic range, as well as the algorithm's accuracy requirements for the intermediate results, select the appropriate DSP. In addition, the floating-point DSP is easier to program. In the fixed-point DSP programming, the programmer must constantly adjust the PQ value of the intermediate result. In fact, the intermediate result is constantly adjusted and rounded.
Real-value operations can be added directly to hardware operations through code, and fixed-point components must be software-indirect to perform real-number operations. This increases algorithm instructions and extends development time.
On the whole, fixed-point DSPs have advantages in cost and floating-point DSPs are better in ease of use.
Since the birth of DSP chips, DSP chips have been rapidly developed. On the one hand, thanks to the development of integrated circuits, on the other hand, it benefits from the huge market. In just over a decade, DSP chips have been widely used in signal processing, communications, and many other fields.
For DSP chip development status and how DSP and other products with application solutions Hikvision Huang Tian made the following observations: DSP chip has been in the direction of specialization, diversification, the market division of various manufacturers more and more Fine, the difference is also growing. In addition, simple DSP chips have been rare, and more is that DSP chips are integrated with other processing cores to form a highly integrated and targeted SOC, which not only greatly reduces board space, but also brings about The overall advantages of power consumption, cost, and development cycle have driven the development of the industry and improved product performance.
The advantage of DSP lies in flexible algorithm integration, which can provide products with powerful performance and flexible customization. The same product can implement different solutions for different customer needs. In order to improve the competitiveness of the products, manufacturers will make full use of algorithms in algorithms, algorithms will become more and more complex, but the stability of the algorithm, the power consumption of the products, and the development cycle will become uncontrollable risks. The DSP algorithm is not a pile of theoretical formulas, but is a sophisticated software that closely integrates with the specific features of the DSP chip used.
Development Status and Application of DSP Chips
These factors need to be fully considered when designing products. Do not blindly adopt so-called advanced algorithms and high-performance DSPs for some gimmick functions. Instead, start with user requirements and find the best combination of algorithms and DSPs. In the product solution, the algorithm and the DSP are the core. This combination determines that, with other processing chips and peripherals, an efficient product solution can be formed.
At present, the use of DSP technology from the military to civilian use, from aerospace to production and life, are increasingly using DSP. DSP technology is mainly used for radar and sonar signal processing in aerospace; in communications, it is mainly used for signal transmission of mobile phones, IP phones (Voiceover IP), ADSL, and HFC; in terms of control, it is mainly used for motor control, Optical drives and hard disk drives; for test/measurement, mainly used for virtual instruments, automatic test systems, medical diagnostics, etc.; in terms of electronic entertainment, mainly for high-definition television (HDTV), set-top box (STB), AC-3, Home theater, DVD and other applications;
In image/graphics, it is mainly used for two-dimensional and three-dimensional graphics processing, image compression and transmission, image enhancement, animation, etc.; as well as digital cameras, web cameras, and so on, all apply DSP technology. At the same time, SOC chip systems, wireless applications, and embedded DSPs are all directions and trends of DSP development in the future. It can be said that without DSP, there will be no access to the Internet, no multimedia, and no wireless communication. Therefore, DSP will still be the technological driver of the entire semiconductor industry. Nowadays, DSP applications continue to expand, and its coverage includes broadband Internet access services, the development of next-generation wireless communication systems, digital consumer electronics markets, and automotive electronics market development.
DSP chip classification
DSP chips specifically designed for different algorithms can be divided into three basic categories: basic characteristics, data formats, and applications.
According to the basic characteristics, according to the DSP chip operating clock and the type of instruction to be classified. If the DSP chip can work normally at any frequency within a certain clock frequency range, there is no performance degradation except for the change in calculation speed. Such a DSP chip is generally called a static DSP chip. If there are two or more kinds of DSP chips, their instruction set and the corresponding machine code machine pin structure are compatible with each other, this type of DSP chip is called a consistent DSP chip.
According to the data format, this is based on the data format of the DSP chip. The DSP chip whose data operates in a fixed-point format is called a fixed-point DSP chip. The work in floating-point format is called a DSP chip. The floating-point format used by different floating-point DSP chips is not exactly the same, some DSP chips use a custom floating-point format, and some DSP chips use the IEEE standard floating-point format.
According to the purpose, it can be divided into general-purpose DSP chips and special-purpose DSP chips. The general-purpose DSP chip is suitable for ordinary DSP applications, such as TI's series of 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.
Conclusion
With the rapid development of the security industry, especially the blowout of high-definition and intelligent demand, the entire industry has fully entered the digital era. As the core of digital security products, security chips have become a huge industry, so more different algorithms are needed. . For the ever-increasing algorithm requirements, the processing power of the DSP is never enough. Algorithm developers need to constantly balance the performance metrics and the processing power of the DSP. It is to make one cycle less, or to allow occasional frame loss, similar to this. If you can't make a choice, you need to further optimize the algorithm, either on the algorithm's architecture or at the assembly code level. On-chip cache is an important factor that affects the performance of the algorithm. DDR bandwidth is often system-level.
Due to the DSP's ability to calculate and not be good at logic processing, especially network protocols and database management, DSPs are often used in conjunction with other chips. Among the various development projects in the field of security, DSP+ARM is the most popular. ARM is responsible for network, storage, and peripheral management. DSP is responsible for image and audio processing and encoding and decoding. At present, DSP+ARM is basically integrated into an SOC. However, for software development, DSP and ARM are also separate. In areas where higher processing performance is required, multi-chip DSPs are often used together for processing. These are also Hikvision Huang Tian, ​​a profound experience of many years of industry senior technical people.
Led Linear Lighting,Hotel Led Linear Batten Light,Led Flat Tube Lights,Ceiling Linear Light
Shen zhen SH LED Technology Co.,Ltd , http://www.pixellightsolution.com