Digital Signal Processor is one of the milestones on the path taken by the computing elements in electronics. When the regular microprocessors and microcontrollers were struggling to handle complex calculations, DSP devices came into picture to rescue the designers from the pain in developing tough algorithms for their applications. In fact the DSP devices have acquired power either to make or kill the applications in many fields.
During the days of 8031 Derivatives, Motorola, Rockwell, Microchip, Zilog parts, when designers were struggling to handle complex mathematical functions, DSP devices helped them manage their applications relatively easy. In many cases, where real time performance were expected, these devices saved the designers from frustration. It is obvious that with given enough time, any computing element can execute the assigned task. But real time systems demand immediate results, most of the time at the blink of eyes, to complete their working and this demands more processing power from the computing elements within the allotted time. But having more processing power is only half the story and the processor should have proper architecture to get more data and code simultaneously to manage fast incoming signals producing the new required output signals with the same rate. The architecture has to be designed exclusively for this purpose with special features. Then the processor should have power to finish requiring supervisory functions to complete the applications.
But the arrival of DSP devices enabled designers undertake many Signal processing applications in audio, video, communication, instrumentation and control systems with ease and confidence. Many stand-alone equipments producing fast real time results came into existence just because of these devices. Naturally the semiconductor major, Texas Instruments, became more popular by launching these devices first and soon many others also followed the suit with their offerings. When TI introduced first generation DSP parts with 5 MIPS capability, everybody was surprised,because at that time, only with mainframe computers, users could get equivalent power. TI went on to create an ecosystem for these parts as a part of their marketing, which was first of its kind that time and then created history.
The DSP devices slowly evolved into an exclusive niche segment within electronics when other semiconductor majors joined with their parts. High resolution Floating point architecture came into picture to handle much complex applications. Even TI launched an advanced device with eight DSP cores working together inside of the TMS320C8xx family devices.
Let us see what makes the DSP devices as powerful and impressive.
1. Hardware Multiplier. All DSP devices come with multiplying capability implemented in hardware. You shouldn’t combine instructions to get this multiplying done like other microprocessors. A simple and single instruction created in the hardware should do the job.
2. Hardware MAC facility. The Multiply and Accumulating facility, MAC, is the basic operator in signal processing. All mathematical equations, when stripped down to their basics, come to multiplying two variables and adding that result to a previously calculated data in the loop. At the first glance the presence of this instruction indicates the controller as suitable for math-intensive operations.
3. Harvard Architecture. This is the main design aspect of DSP parts, enabling the core get both program code and the required data simultaneously to speed up the processing. In latest DSP parts, core gets more data and code in every access.
4. Pipeline architecture. Pipeline facility is being added to the high speed processors to speed up their executions. During program execution, multi stage pipeline keep more than one instructions in the pipe in various dissected states to speed up their executions. After initialisation, because of this pipeline, the processor starts executing more than one instructions in every clock cycle to finish calculation intensive operations.
5. Onchip Memory. DSP processors keep both program code and the required data within the device itself. During the execution, the processors don’t waste time accessing code and data from the external storage. By design, all required content is kept inside or collected in advance and ready for the execution.
6. Single cycle execution. All the above features enable the processor execute more than one program instructions in every clock cycle. In fact, this indicates the performance of the target processor.
7. To make all the above work in sync, DSP processors have other support features. They have special addressing modes to access variables faster with minimum time to speed up processing in the loop. All processors come with minimum of 32bit width to avoid overflow and truncated errors during algorithm execution. Circular buffering is also being used to manage loop operations.
As you know, the complete DSP processing involves the data acquisition from the input sensors through analog to digital converters as digital data which go through the processing which creates the required results in digital form and then pushed into digital to analog converters producing analog output signals.
A typical DSP flow contains input sensors monitoring the target environment or measuring a target signal using analog to digital converters in the first stage. In general, AD converters with 8-16 bits data range are widely used to sample the raw analog signals. Then processing core proceeds further with the sampled digital data and do all the required calculations on the input data with the predefined coefficient values as needed by the chosen algorithm. Output of this step is the required signal in the digital form. In the last step, waiting DA converter changes this digital data into the exact analog waveform as the final output to interact with the real world. These steps are inevitable in any signal processing and the DSP processors can comfortably handle all this at a faster rate than other choices.
When the real time processing is required, all these steps should be executed for every input sample without fail. In fact, the DSP processors should complete all the operations in the fraction of the sample time, coz they have to complete other supervisory tasks like storing results for further operations, displaying results to the users’ convenience and etc. A powerful processor should keep more than 50% of the sample time for other application related tasks to make the application a successful one. In communication field, the real time requirements should be met without any compromise. Even when the real time processing is not a big requirement as in medical electronics, where the need is to extract the exact waveform out of noisy input signals, DSP processors can do a better job than other options in minimum time. In industrial controls, handling the real time control algorithms is very much possible using DSP processors.
When the designers have understood the potential of the DSP processors, everybody started using the same in their applications. Many new products, which were not possible before, also came into existence thanks to the DSP processors.
New Microcontrollers with DSP capability
During two decades, since the birth of DSP processors slowly penetrated into many segments without much difficulty. Major semiconductor manufacturers TI, Analog Devices, Motorola launched many versions of these devices to expand the market. Fixed point and Floating point devices started proliferating the market to capture more customers than before. Even though the cost DSP devices were higher than top microcontrollers of that time, designers preferred them because of convenience and ease of using them. Naturally the trend which favored the DSP parts made competitive micon majors adapt their approach in their own products to meet the challenges posed by DSP. Slowly the micons acquired more teeth to recapture the customers who started aligning with DSP parts.
As the first step, top micons started getting built-in hardware multipliers to speed up the processing.
MAC instructions also became part of these micons’ power arsenal.
Harvard architecture was modified to access more variables and code in every instruction cycle.
Pipeline naturally was fused into micon design.
Micon majors worked on their instruction set to get more number of instructions executed in single clock cycle speeding up performance. Most frequently used instructions are optimised for single cycle executions.
Operating frequency of the micons also increased gradually to support all the above as the manufacturing technology improved.
Along the way, micon manufacturers started using the inherent advantages of the micons: peripherals. The top micons come with a range of data converters, versatile timer functions and enough built in memory space to take up new challenges. High speed data transfer features within the device like simultaneous sampling, DMA supported the above and pushed the performance higher.
By this time, all designers found that only with micon manufacturers, they get affordable pricing. Most of the time, they could use the existing software tools to manage the top performing devices to complete their applications. Learning curve in the process is very much simplified and all device manufacturers support their devices with suitable DSP code libraries.
Digital Signal Controller Vs Advanced Micons
Now it was the turn of DSP part manufacturers feel the heat on their strategy and decided to take on the competition with new devices. DSP parts started having data converters inside, more peripherals are also served within the parts and were priced attractively. Manufacturers who were selling multi thousand tools offered their tools matching micon manufacturers.
One thing that went wrong for the DSP devices were their interrupt structures. They didn’t have versatility of micons interrupt management. May be the device architecture don’t allow any faster interrupt management. Personally, when we tried to use the TI’s new devices in few of our applications, we found their interrupt features as not useful and went back to Renesas Microcontrollers to finish the designs.
Micons manufacturers recaptured segments catering to industrial controllers in the first step. New features like built in full scale LCD interfacing and more memory helped the designers with these micons. Still many can manage their applications comfortably creating the real time bandwidth in the range of 100-200Khz using the micons. More designers who had taken up DSP devices turned back to micons. Industrial controllers were the first segment DSP manufacturers lost to their micons counterparts.
ARM Parts VS DSP Devices VS Microcontrollers
In 1985, a British company launched 32 bit ARM microprocessors into the market. Operating at 3MHz, it produced 3 MIPS performance. With this start, the ARM, started enhancing their range. ARM is still fabless company, selling only design of ARM core to the semiconductor manufacturers, who in turn add their peripheral functions to complete the chip design. This facility enabled all the micon majors customise the ARM core to their likings. Almost all semiconductor manufacturers started using the ARM core in their Microcontrollers.
Because of the on going pressure from DSP devices, ARM too announced micon core supporting DSP instructions in the family ARMv5JET. Now many ARM core based Microcontrollers are available from more than 100 semiconductor vendors and in this scenario, only a few micon manufacturers still retain their proprietary micon designs in their portfolio. But they too maintain a range of ARM core based micons to keep customers with them. That’s the reason, in 2013, ARM announced shipping of 50 billion ARM core based micons.
Where is the dedicated DSP now?
Now we don’t see much news on the DSP devices which ruled the electronic fields of 1980s and 90s. Personally I used to watch advertisements issued by various DSP part manufacturers in 80s claiming special features of their devices. New generation Microcontrollers literally pushed these DSP parts into underground. In fact, all mobile phones use the DSP parts in their designs, but as a sub module in the big SOC device designed exclusively for the phones. Now we don’t see any news on the DSP parts like before. Now they cater only to higher end communication and entertainment as a part of exclusively designed bigger SOC devices. Even though these DSP devices sitting in our pockets, we really don’t know much about them.
Apart from mobile phones, high end media players, entertainment systems, DSP devices literally lost all the gained market share to Microcontrollers. The real winners are the actual designers who have got the best of both worlds helping them to undertake advanced applications than before.