Two new high-precision control digital signal processors (DSPs) unveiled today from Texas Instruments Incorporated (TI) reduce development time from hours to minutes while delivering up to 12x performance of any existing programmable DSP controller. The new devices, the industry's first 32-bit control DSPs with on-board Flash memory and performance up to 150 million instructions per second (MIPS), target industrial automation, optical networking and automotive control applications.

Optimized for Control Applications

The new TMS320F2810 DSP and the TMS320F2812 DSP are the first catalog devices based on TI's code compatible TMS320C28x DSP core, which is designed exclusively for control applications and is extensible up to 400 MIPS. With an innovative unified architecture that combines the best of general-purpose processors and DSP capabilities, the C28x devices allow both the system and math code to be developed completely in C/C++, reducing development time from hours to minutes. Hardware and software features that help simplify development include:

- 150 MIPs
- 32-bit processing
- Efficient C/C++ compiler
- IQmath library
- Fast and flexible interrupt management
- Control optimized peripherals
- Multiple Standard Communications ports
- Up to 128Kw of on-chip Flash memory
- Application-specific packaging
- Code compatibility with TMS320C24x generation
- eXpressDSP

TI's TMS320C28x DSP delivers the industry's most integrated DSP controller while improving system reliability, flexibility and intelligence for tunable laser applications in the optical networking space. A version of the F2812 DSP that is packaged in a 179-lead microStar ball grid array (BGA) meets the tighter space requirements of optical networking. The TMS320C28x DSP is the latest addition to TI's portfolio of products designed to enable the intelligent optical networks of today and tomorrow.

The new F2810 and F2812 DSPs offer a complete system-on-a-chip for high-precision control applications leading to simpler more cost effective designs. The F2812 DSPs integrate 128 kilowords (kW) of Flash memory and the F2810 DSP 64 kW for reprogramming during development and in-field software updates. TI’s Flash acceleration technology allows code to be executed out of Flash at 110-120 MIPs while time-critical code requiring 150 MIPs of performance can be executed directly out of the 18kW of on-chip RAM. In addition, the F2812 DSPs offer an external memory interface with an address reach of one megawords (Mwords) for systems requiring a larger memory model.

Integration of control peripherals and communication ports in the new C28x DSP devices provides cost-efficiency and ease of design for controller applications. Optimized event managers include flexible pulse-width modulation (PWM) generators, programmable general-purpose timers and glueless capture and encoder interfaces. A 12-bit analog-to-digital converter (ADC) capable of two channel conversions at 60 nanosecond (ns) per channel and dual sample-hold provides critical support in the completion of analog control loops. Multiple standard communication ports provide simple communications interfaces to hosts, test equipment, displays and other components. A Control Area Network (CAN) controller that is version 2.0B-compliant provides a glueless interface to CAN networks for a variety of industrial, automotive and backplane applications.

The C28x DSPs and C/C++ compiler were designed together to ensure maximum efficiency of compiled code allowing designers to develop control applications entirely in a high-level language. A tight, 16-bit instruction set generates superior code density while providing 32-bit precision. Code developed for C28x DSPs is code compatible with the C24x DSP generation, protecting developers software investment as they migrate their designs allowing them to take advantage of TI's newest DSP generation.

In addition to the efficient C/C++ compiler, the C28x DSPs are equipped with 32-bit math capabilities as well as a unique math library allowing designers to port floating-point algorithms to a fixed-point machine in a matter of minutes. With the conversion of math algorithms in traditional 16-bit fixed-point machines being a laborious and time-consuming task, TI's C28x DSPs give designers the math code efficiency of DSPs with the ease and system code efficiency of general purpose processors making development of control applications faster and easier.

Texas Instruments Incorporated, Semiconductor Group, SC-01130, Literature Response Center, P.O. Box 954, Santa Clarita, CA 91380. Phone: 800-477-8924, ext. 4500

Even though the TMS320C28x is a DSP, I decided to review it because of its integrated analog-to-digital converter and its use for analog control applications. The chip integrates a 12-bit successive approximation register (SAR) analog-to-digital converter, has a 32-bit register, and uses a C/C++ compiler to take advantage of some high-precision capabilities. For you design engineers with applications in industrial automation, automotive, optical networking, power conversion, and even office equipment, this chip series may be just what you are looking for because it offers real-time control.

The idea of control is big with these DSPs. It provides 32-bit precision with a 64-bit accumulator, allowing 64-bit multiplies. TI also focused on the control code. There are two areas of control to consider - the system code and general purpose code, and each has different requirements. TI tried to hit the sweet spot and allow the designer to use a single architecture to support both types of code. This would allow control to be implemented by a single processor, instead of the usual two. Reducing the control to one chip is cheaper, right? But how do you accomplish the different types of control using just one processor? TI made a chip that has more MIPS - up from the typical 40 to 110 MIPS, added 32-bit control, and then provided a math library to help aid in application development as well as C/C++ for system coding. They didn't stop there, the company added a real-time Joint Test Action Group (JTAG) interface for quick debugging. The real-time JTAG port allows access to every register and memory location through hardware, while the processor is running. This is especially helpful for engineers using RF devices.

The 12-bit ADC can make 200ns single conversions or 16.7 MSPS, which provides a continuous throughput. It also has a dual sample-and-hold that is very useful if you want to sample two phases of a current, simultaneously. The integrated ADCs have built-in 16-bit FIFOs, which means they can sequence up to 16 conversions without CPU intervention and even sample an entire waveform. Pretty cool.

The 2810 is targeted for industrial drives, electronic power steering, and integrated starter/alternators. The 2812 has twice the flash memory on-chip as the 2810 and has an external memory interface. Appropriate applications for the 2812 include robotic systems, uniterruptible power supplies, and optical networking.

TI realized that just the silicon wasn't enough to support all these control applications. They realized that software was also an important piece of the puzzle. They provide foundation software with an operating system, and an IQMath chip support library (the peripheral drivers). Layered on top of that is an application specific library provided by TI. For example, you could use the TI digital motor control library that allows you to develop an entire system.

The IQMath library is interesting and worth looking at because it allows you to change from floating point to fixed point math. Most math algorithms that are time-critical start in a floating point format. However, performance is impacted when you run the floating point algorithms on a fixed-point machine. The typical way to deal with this meant you had to get the algorithm running in floating point and then port it over to fixed point to improve performance. The changing to fixed point takes time and becomes convoluted and hard to maintain. TI decided that they needed to help engineers with a library that allows them to change from floating to fixed point math. So, using a high-level language, all you have to do is replace the calls for a floating point library function call with the IQMath function call - and you're done.

Samples of the new TMS320C28x DSPs are scheduled for availability during the first quarter of 2002, with volume production scheduled to follow in the second quarter. The F2810 DSP will be offered in a 128-lead LQFP, and the F2812 DSP in a 176-lead LQFP which is well-suited to the production needs of industrial, automotive, commercial and consumer applications. A version of the F2812 DSP that is packaged in a 179-lead microStar ball grid array (BGA) meets the tighter space requirements of optical networking. Planned pricing per unit in quantities of 10,000 is $17.95 for the F2810 DSP and $22.95 for each version of the F2812 DSP.