Hello, i'm quai。
In the field of embedded development, dsp (digital signal production)Digital signal processingA microprocessor for high-speed digital signal processing。
Although i'm mainly involved in embedded linux and single-piece development, i've also been exposed to a number of scenarios that require dsp involvement, such as audio processingElectrical controlImage recognition, etc。
Today we will talk about the selection and application of the dsp, with a view to informing our friends who need it。
1. Dsp foundation concepts and characteristics 1. 1
Dsp is a microprocessor specially designed for digital signal processing。
Dsps have a clear advantage over generic mmus (e. G. Stm32) when dealing with a large number of mathematical operations。
Its hardware architecture optimizes operations such as multiplication, multiplication and so forth, usually using the harvard structure (segregation of programs and data repositories) to complete the implementation of multiple instructions within a clock cycle。
When i was working on the auto electronics project early in the day, there was a need for engine noise to be proactive。
The cpu occupancy rate was often over 90 per cent when the normal arm cortex-m4 audio data were processed, and was very poor in real time。
The cpu occupancy rate for the same algorithm dropped to around 30 per cent after the c2000 series of dsps, which were replaced by tti, with immediate results。
1. 2 core features of the dsp
Dsp has several notable features compared to regular mmus。
The first is a dedicated hardware multiplication, which can be performed within a single clock cycle。
The second is the multi-line structure, which allows simultaneous access to both the program and the data repository。
The third is a dedicated site search model, such as a circular address, a reverse location, etc., which is very useful in algorithms such as fft。
The fourth is the abundance of external interfaces, especially high-speed adc, pwm, which are well suited to electrical control and audio processing。
To give a practical example, a large number of trigonometric functions and coordinate conversions are required for the electrical foc (magnetic field orientation control)。
If the stm32 f4 series were to be used, it would take dozens of microseconds to process a full foc cycle, even if the fpu (floating point unit) had been activated。
The c2000 series of dsps using ti can be performed in a few microseconds, which is essential for high-speed electrical control。
2. Mainstream dsp manufacturers and product series 2. 1 texas instruments (ti)
Tti is the head of the dsp, and the product line is very rich。
The c2000 series is primarily oriented towards real-time control applications, particularly in the areas of electrical control and digital power。
This series of dsps is a collection of high-precision pwms, fast-track adcs, etc., that are well suited to industrial controls。
The c5,000 series focuses on low-capacity applications, often for portable audio equipment, hearing aids, etc。
The c6000 series is a high-performance dsp for use in locations requiring extensive data processing, such as communications bases, medical imagery, etc。
When i was doing the auto electronics project, i used tts320f28335, a typical c2000 number。
It has a main frequency of 150 mhz, with 12-bit adcs and a conversion time of only 80 ns, which is well suited for fast current sampling。
We used it to power the system, to collect electric currents in real time and to control the foc, a fully competent dsp。
2. 2 adi
The sharc series dsp of adi is well known in the field of audio processing。
Sharc dsp uses an ultra-platform structure, with a strong floating-point computing capability, which is particularly suitable for applications such as specialized audio equipment, acoustic processing, etc。
The blackfin series is a fixed-point dsp, which is less expensive and often used in areas such as video surveillance, image processing and so on。
I have a friend who's working on specialized sound equipment, and they use adi's adsp-214889, a double nuclear shacc dsp with a primary frequency of 450 mhz, capable of simultaneously processing multiple audio signals to achieve complex audio effects such as mixing, balancing, compression, etc。
2. 3 nxp and st
The i. Mx rt series of nxp, although not purely dsp, has been assembled as a dsp co-processor to handle some medium complexity signal processing tasks。
Stm32h7 is similar, with the main core being arm cortex-m7, but performance is already adequate for many dsp tasks。
In practical projects, if signal processing needs are not particularly complex, i usually give priority to stm 32h7。
For example, as a simple audio filter, stm32h7 is fully adequate and the development of a tool chain is more mature and easier to debug。
3. Key element 3. 1 of the dsp selection
Select the dsp first to specify the operational performance requirements。
This includes precision (fix or float), operating speed (mps or mflops), storage capacity, etc。
Targeted dsps are low-cost and low-utility, but they are limited in precision and suitable for settings that require less precision。
Floating point dsp is highly accurate and programmable, but cost and cost are relatively high。
For example, if a simple digital filter has only a few khz sampling rates and data accuracy is not required, then the dsp is sufficient。
However, if the sound is decoded, the audio needs to be processed at 44. 1 khz or higher, and if the sound is to be preserved, the floating point dsp is required。
In one of my projects, an 8-step ir filter with a sampling rate of 10 khz is required。
We initially chose the fixed point dsp, but we found that quantitative errors caused the filter to be unstable。
Later on, the problem was solved by turning into a floating point dsp。
This lesson tells me that the need for precision in the calculation must be fully assessed in the selection。
3. 2 external interface requirements
The external interface of the dsp is also an important consideration in the selection。
For electrical control applications, high precision pwm, fast adc, encoder interface etc. Are required。
For audio applications, audio interfaces such as i2s and mcasp are required. For communications applications, high-speed chains, ethernets, etc. Are required。
In the c2000 series of tti, for example, it integrates epwm modules that produce high-precision pwm waveforms, and death zone time to nanoseconds。
This is important for electrical control and digital power。
It also integrates the eqep module, which allows direct connection to the incremental encoder, hardware decodement, and does not occupy cpu resources。
I used the epwm and eqep modules of tms 320f2806 when i was working on a three-stage unbrusher control project。
The epwm can generate six circuits to complement pwm, protect the dead zone and drive the three reversers directly。
Eqep can read the location and speed of the encoder and achieve closed loop control。
These hardware peripherals greatly simplify software development and improve system reliability。
3. 3 development of tools and ecosystems
It is also important to develop the ease of use of tools and the level of integrity of ecosystems。
Ti's code composer studio (cs) is a more mature industry dsp development environment that supports the c/c++ programming and is a set of debuggers, performance analysis tools, etc。
Adi's crosscore embededed stadio is similar。
In addition, consideration will be given to the availability of rich library functions and illustrative codes。
Ti provided a controsuite, containing a large number of examples of applications such as electrical controls and digital power。
Adi has similar resources. These resources can significantly reduce the development cycle。
I personally prefer the development environment of tti because its documentation is very detailed and the community is active。
When problems are encountered, answers are usually found in the ti e2e forum。
Also, tti provides high-quality library functions, such as the iqmath library, which can simulate float-point operations using fixed-point operations, both ensuring precision and increasing speed。
3. 4 cost and supply stability
Cost is a factor that must be taken into account in a commercial project。
Dsp prices range from several united states dollars to several hundred united states dollars, and the appropriate model is selected according to the project budget。
At the same time, consideration should be given to the stability of supply, especially in the case of volume-producing projects, the choice of a long and stable supply pattern。
Supply stability is particularly important in the area of automobile electronics. Automobile products usually have a life cycle of more than 10 years, so the dsp we choose must ensure long-term supply。
Ti's c2000 series is doing well in this regard, with many models having been available for more than a decade and promising to continue。
4. Typical dsp application scenario 4. 1 electric control
Electric control is one of the most typical applications of the dsp。
Modern electromechanical control algorithms, such as coc, non-sensor control, etc., require a large amount of mathematical operation。
The high-speed computing capability of the dsp and the abundance of external interfaces make it an ideal option for electrical control。
For example, the foc algorithm requires a series of calculations such as clarke, park, pi control, antipark, and svpwm。
These operations involve a large number of triangular functions and matrix operations。
It is difficult to complete all of the calculations in a pwm cycle with an ordinary mmu. And with the dsp, it's easy to control dozens of khz。
The following is an example of a simplified coc control code (false code):
I'm sorryNtrol (void)
other organiser
/ / read 3-phase current
fload ia = adc readcurrent a();
fload ib = adc readcurrent b();
fload ic = adc readcurrent c();
/ clarke transformation: abc coordinate system converted to alpha-beta coordinates yes
fload i alpha = ia;
fload i beta = (ia + 2*ib) / sqrt(3);
// read rotor position
float theta = encoder getangle();
/ park conversion: conversion of alpha-beta coordinates to dq coordinates yes
fload id = i alpha * cos (theta) + i beta * sin (theta);
fload iq = -i alpha * sin (theta) + i beta * cos (theta);
/ pi control
float vd = pi controller d (id ref-id);
float vq = pi controller q (iq ref-iq);
// anti-park transformation: dq coordinates converted to alpha-beta coordinates yes
float v alpha = vd * cos (theta) - vq * sin (theta);
fload v beta = vd * sin (theta) + vq * cos (theta);
/ svpwm modem
svpwm modulation (v alpha, v beta);
♪ i'm sorry ♪In the actual dsp code, these trigonometric functions can be performed at a very fast pace by tabulation or hardware acceleration。
4. 2 audio processing
Audio processing is another important application of the dsp。
These include audio decoding, sound processing, noise reduction, echo elimination, etc。
These applications require the processing of a large amount of audio data and are highly demanding for real time。
I've been involved in a car-borne acoustic project that requires active noise reduction。
The rationale is to collect environmental noise through microphones, which are processed by dsp and produce backsonic waves to offset the noise。
This process needs to be completed in a few milliseconds, otherwise the impact of noise reduction will be significantly reduced。
Typical algorithms for audio processing include fir filters, ir filters, ffts, etc。
Dsp is very supportive of these algorithms。
For example, the tti c5000 series, dedicated to the optimization of audio applications, provides a dedicated audio processing library。
The following is a simple example of the fir filter code:
#definine filter length64
float fir coefs
/ / filter coefficient
0. 001, 0. 002, 0. 003, ...
};
other organiser
int buffer index = 0;
float fir filter
other organiser
float output = 0;
/ update buffer zone
= input;
buffer index = (buffer index + 1) % filter length;
volume operations
for (int i = 0; i)On the dsp, this cycle can be optimized through hardware acceleration or simd commands, significantly increasing the speed of operation。
4. 3 image processing
Image processing is also an important application of the dsp。
These include image enhancement, edge detection, image compression, etc. These applications require processing a large amount of pixel data, which is very large。
In industrial visual detection projects, real-time processing of images collected by cameras is often required。
For example, detection of product defects, identification of 2d codes, etc. These tasks would be handled very slowly if they were handled with regular mmus。
And with dsp, real-time processing is possible。
Typical algorithms for image processing include volume, morphology, hof conversion, etc。
These algorithms involve a large number of matrix operations and are well suited for dsp processing。
4. 4 communications signal processing
In the area of communications, dsps are used for modems, channel decoding, signal detection, etc。
For example, in the 4g/5g base station, there is a need to process a large number of wireless signals and perform ofdm modems, channel estimates, equilibrium, etc。
All this needs high-performance dsp to do。
Although i did not have a direct communication project, i also had contact with the signal processing of the can bus in the automobile electronics project。
Although not as complex as wireless communications, precise time-series control is required for the bit-series detection, error detection, etc. Of the can bus。
5. Accountancy 5. 1 techniques developed by dsp
If the dsp is used, special attention needs to be paid to numerical precision and spill problems。
Targeting operations require programmers to manage decimal places manually, with a slight lack of attention leading to a loss of precision or spillage。
The iqmath library provided by ti is a good tool for the simulation of floating point calculations using integer calculations, which both guarantee precision and increase speed。
When using the iqmath library, you need to define a q format for data, such as q15 for 1 bit sign and 15 decimal places。
// example using iqmath library
#include "iqmat lib. H"
iq value1 = iq(1. 5); // defines the number in an iq format with a value of 1. 5
iq value2 = iq (2. 3);
iq result = iqmpy (value1, value2); / / iq format multiplication
fload result faat = iqtof (result); / / convert to floating point5. 2 optimizing code performance
In dsp development, code optimization is important. The hardware features of the dsp, such as hardware multipliers, circular buffer zones, dma, etc., should be fully utilized。
Optimization options for the compiler are also set reasonably, and it is usually recommended to use either o2 or o3。
Key codes can be developed using either the compiler language or the internal function of the compiler。
For example, ti's dsp supports many internal functions that can be directly mapped to hardware commands。
/ / example with built-in functions
#includeInt a = 10, b = 20;
int sum = add2 (a, b); / / using built-in functions to add to 5. 3 real-time assurance
The dsp application is usually very demanding for real time and requires assurance that the operation is completed within a specified time frame。
This requires programmers to prioritize their tasks rationally, avoid interruptions of embedded depths and rationalize the use of dma to reduce the burden on cpus。
When i did the electrical control project, the coc control task was of the highest priority and had to be completed within each pwm cycle。
Other tasks, such as communications, displays, etc., are of low priority and can be performed during free time。
This ensures the real-time nature of the control algorithm。
5. 4 debugging techniques
Dsp debugging is relatively complex, especially for applications with high real-time requirements。
Ccs provides many debugging tools such as real-time observation windows, graphic displays, etc。
It is easy to observe variable changes while the program is running。
Personally, i prefer to use the ccs 's graph function to draw waves in real time。
When debugging the foc algorithm, for example, current wave shapes, speed curves, etc. Can be observed in real time and control effects can be seen in visual terms。
6. Summary and outlook
The dsp, as a dedicated signal processor, has an irreplaceable role in areas such as electrical control, audio processing and image processing。
The selection of a suitable dsp requires a combination of operational performance, external interfaces, development of tools, costs, etc。
With the development of technology, the boundaries between dsp and mmu have become increasingly blurred。
Many high-performance mmus, such as stm32h7, i. Mx rt series, are now well equipped for signal processing。
These mmus are fully capable of medium complexity applications。
However, dedicated dsps remain the best option for high performance, high real-time applications。
From my own experience, if the project budget is adequate and performance is high, it is recommended that a dedicated dsp be chosen。
High performance mmus can be considered if budgets are limited or signal processing needs are not particularly complex。
Whatever option is chosen, needs are fully assessed and technical validation is made to ensure the correctness of the selection。
Finally, the development of the dsp is relatively complex and requires a solid mathematical base and extensive practical experience。
It is suggested that beginners start with simple applications and move in further. Producers such as tti, adi and others provide a large amount of learning resources and illustrative codes that can be used more effectively。
It is hoped that this article will help you by making less of a turn in the dsp selection and application。
