Welcome to the Blogs section of the TI E2E Community! Ask questions, share knowledge, explore ideas, and help solve problems with fellow engineers on TI’s Engineer-to-Engineer (E2E) Community
Gene FrantzTI Principal Fellow, Futurist and Business Development Manager, DSP
I am currently running a series looking at where technology will be in 2020. I hope you are enjoying it! I wanted to break into that series for a minute to comment on a paper I just finished reading from Robert Cravotta (EDN Magazine) entitled: “The evolving landscape of digital signal processing.” It’s probably a good time for me to add to Robert’s well written view of where DSP is headed. After reading the article, or on first glance, it may appear that DSP is dead - which is why I feel it is important for me to add to Robert’s paper. And, since this is the message [DSP is dead] I began driving inside of TI several years ago, I thought it would be good to continue that discussion in this more public forum. When we introduced our first DSP device (TMS32010) in 1982 there was little in the industry that could match its performance. But, it became apparent to me in the late 1990’s that a new era of signal processing was emerging. Most microprocessors, microcontrollers and FPGAs were beginning to have enough performance to be reasonable alternatives to a DSP, at least for relatively simple DSP algorithms. I also noticed that for many of the new opportunities the general purpose DSP was not adequate. We, at TI, decided to address this head on and put together a small team that began looking at a new concept to add an accelerator to enhance the performance of the DSP. I called the accelerators “semi-programmable DSPs” since within a very narrow application segment the accelerator was infinitely programmable. For example, the 16-bit hardware multiplier (our first accelerator – used on the TMS32010) was infinitely programmable as long as what you wanted to do was 16 X 16 bit multiply. One of our senior managers thought it best that we not use these accelerators as the market might figure out that the DSP could be replaced with an ARM plus accelerator combination. I countered with the fact that the TMS32010 was nothing more than a microcontroller with an accelerator (i.e., hardware multiplier). I even went as far as to give my crude internal explanation that “DSP, as a product, is dead.” But, even more importantly, “DSP as a technology would enable every product we made in the future.” In fact our DSP product line has reflected that notion for well over a decade now. Whereas the TMS32010 was a DSP for all markets, through the years our DSP product line has reflected a different story. The TMS320C2xx family became processors for the motor control market. The TMS320C5xx family became processors for the handheld cell phone market, later augmented by the OMAP family of processors. The TMS320C6xx family became the processor for the video processing and communications infrastructure markets, later to be augmented by our Davinci product line. So, here we are well into the 21st century and we are asking ourselves if DSP is dead. This is such unfortunate timing because when you really look at this, the contratry is true…the use of DSP is expanding more rapidly than ever before! We sometimes forget were we are in the world of DSP. Our: -Phone system is digital -Audio is digital, from the handheld to the theatre -Pictures are digital -TV is digital -Control is digital All of this technology revolution is the result of DSP technology and high performance math engines. And, this will be the basis of the next wave of innovation. That’s right, it is only beginning and not ending. It’s an exciting time, the caterpillar is about to become a butterfly. I would love to hear your thoughts on the topic. Leave your comments here. Gene
I am currently running a series looking at where technology will be in 2020. I hope you are enjoying it! I wanted to break into that series for a minute to comment on a paper I just finished reading from Robert Cravotta (EDN Magazine) entitled: “The evolving landscape of digital signal processing.”
It’s probably a good time for me to add to Robert’s well written view of where DSP is headed. After reading the article, or on first glance, it may appear that DSP is dead - which is why I feel it is important for me to add to Robert’s paper. And, since this is the message [DSP is dead] I began driving inside of TI several years ago, I thought it would be good to continue that discussion in this more public forum.
When we introduced our first DSP device (TMS32010) in 1982 there was little in the industry that could match its performance. But, it became apparent to me in the late 1990’s that a new era of signal processing was emerging. Most microprocessors, microcontrollers and FPGAs were beginning to have enough performance to be reasonable alternatives to a DSP, at least for relatively simple DSP algorithms. I also noticed that for many of the new opportunities the general purpose DSP was not adequate. We, at TI, decided to address this head on and put together a small team that began looking at a new concept to add an accelerator to enhance the performance of the DSP. I called the accelerators “semi-programmable DSPs” since within a very narrow application segment the accelerator was infinitely programmable. For example, the 16-bit hardware multiplier (our first accelerator – used on the TMS32010) was infinitely programmable as long as what you wanted to do was 16 X 16 bit multiply. One of our senior managers thought it best that we not use these accelerators as the market might figure out that the DSP could be replaced with an ARM plus accelerator combination. I countered with the fact that the TMS32010 was nothing more than a microcontroller with an accelerator (i.e., hardware multiplier). I even went as far as to give my crude internal explanation that “DSP, as a product, is dead.” But, even more importantly, “DSP as a technology would enable every product we made in the future.”
In fact our DSP product line has reflected that notion for well over a decade now. Whereas the TMS32010 was a DSP for all markets, through the years our DSP product line has reflected a different story. The TMS320C2xx family became processors for the motor control market. The TMS320C5xx family became processors for the handheld cell phone market, later augmented by the OMAP family of processors. The TMS320C6xx family became the processor for the video processing and communications infrastructure markets, later to be augmented by our Davinci product line.
So, here we are well into the 21st century and we are asking ourselves if DSP is dead. This is such unfortunate timing because when you really look at this, the contratry is true…the use of DSP is expanding more rapidly than ever before!
We sometimes forget were we are in the world of DSP.
Our:
-Phone system is digital
-Audio is digital, from the handheld to the theatre
-Pictures are digital
-TV is digital
-Control is digital
All of this technology revolution is the result of DSP technology and high performance math engines. And, this will be the basis of the next wave of innovation. That’s right, it is only beginning and not ending. It’s an exciting time, the caterpillar is about to become a butterfly.
I would love to hear your thoughts on the topic. Leave your comments here.
Gene
Hello Gene
I remember it was at DSPFest 2000 or 2002, you took the microphone and exclaimed "DSP is dead!". I was left perplexed and wondering what was cooking at TI...!
It took a few years but finally I have the explaination.
I think many would agree, as you state, that DSP can take many forms and can be implemented on many hardware platforms and architectures, including FPGAs and microcontrollers.
The idea of a DSP accelerator (a simple multiplier or MAC in its most elementary form) is still in vogue, most notably in TI's MSP430 family where a fast multiplier allows a surprizingly high level of DSP performance. However I think that many signal processors (as in "signal processing developers") would agree that DSP takes more than a simple computation accelerator.
To me DSP hardware should include the following three characteristics:
- Facilitate and accelerate signal processing. More generally I should say "facilitate Computationally Intensive Processing, such as Vector Processing. Indeed I think it is difficult to argue a tangible difference between "true-blue" signal processing and other forms of computationally intensive math. Nor should we limit our thinking on the basis of established application fields.
- Insure positive timing control. It is not sufficient in my view that the signal processing hardware increases computation speed. It is also essential that the hardware and architecture make the execution time deterministic. Some "hardware-accelerated" architectures on the market advertise signal processing functions but behave in such a way that computation time is signal-dependant (for instance because multiplications take a drastically different time to execute depending on the values of the arguments). To me this is unacceptable as a legitimate signal processing solution.
- Be low cost... Who needs dedicated signal-processing harware when money (or size, or power consumption) is no object! What completely changed the world 27 years ago was not the development of the first computationally-able CPU, many high-performance CPUs and Vector Processors had been in existance before. It was the development of the first LOW-COST computationally-able CPU.
Having worked with architectures where the multiplier is external to the ALU I must say this is NOT the most comfortable situation for the developer. Having a mutiplier (or a MAC) within the ALU, directly supported by the instruction set, allows a more efficient use because it avoids reentrance and resource sharing problems. Besides if the architecture is really dedicated to Computationally Intensive Processing it should be natural to have a "computationally able" ALU embedded at its core.
Keep-up the good work!
Bruno
Last year I did start working on a portal called dspisopen.com, where I wanted to merge the open source Linux community with DSP world. According to me, w.r.t DSP - there are two different sets of programmers one working with generic processors like ARM using High level OS and others programming the DSP for a specific use case. Both would exist on there own and grown but the challenge for us would be to to integrate these to communities or make then understand the other's need, make them work together.
- More thoughts here khasim.blogspot.com/.../dspisopencom.html
Overall, the job of DSP is to make ARM look better.
I think we should not look at our processor as DSP but a fast processor that can do MAC. That has alwasy been my view of DSP's. A lot of people get scared if you mention DSP's but if you tell them that it is great at doing any functions and you can program using C, then their approach changes. I think once we add an MMU to our devices then we are talking general purpose processor that can do a lot of functions very fast.....
Regards
Mohsen
I want to share my opinion.
It will be good if TI will add some Davinci DSP in DIP trough hole packages. Because I am sure 100% more people will be attracted to it.
Listen to my suggestion, it is better to try to develop new DIP package and see what will happen instead of not try.
I have work with DSP in my MSc in Brunel University and have discover many things trough learning this new platform. I am also in the TI page in february 2009 for my work in Video processing. www.ti.com/.../hof.htm
I am very advaced (can say professional) user in microcontrollers where I used AVR, Cypress, Renesas,ZiLOG and NXP ARM. I am sure there is a space for DSP you just need to make it possible for simple user to start to access to your hardware and software. Because all this development board and Beagle board is nothing, the beauty of electronics is to create some times prototypes with your hand by soldering some components and IC. The user must have the access to simple things first before he will spend money for something faster.
Also look at the facts now for example Cypress moving very smart their C compilers for their PSoC series after HiTECH was obtained by Microship, the Cypress deliver free IDE + C Compiler. So every thing is free like in AVR the WinAVR GCC compiler.
So, my opinion of 8 years of exprirince in the MCU world and 2 years with DSP is that TI must think in the way of free software and DIP package hardware.
If TI think to make money from software this is wrong you already sell your video, audio codecs and etc.
So, at least give IDE and compilers completle free for some of DSPs.
Look in OS it was Windows now is also Linux there :) and it is free, People need free software.
Why you thinking only about huge companies like Nokia which make some of their phones with TI DSP?
Think about simple user, most times this is like domino...