Dr. Finbarr Moynihan, of Analog Devices describes how changingdemands for embedded control have lead to the development of single chip devices based on core competencies in digital signal processing.

In recent years, there has been a major adoption of DSP technology in a variety of embedded control applications. Typical industrial applications include high-performance servo drives in applications like machine tools (milling machines, CNCs), robotics (in automobile assembly lines) and production machines for printing, packaging, textile, weaving, semiconductor wafer handling, bottling, sorting, and other automated production applications. Additional industrial-control applications that require DSP performance include uninterruptible power supplies (UPS), high-end switched-mode power supplies (SMPS), and general-purpose variable-speed drives in cranes, elevators, and process automation equipment.

Common requirements

All of these seemingly disparate applications share a common set of requirements: the controller regulates a switching power-converter based on feedback measurements of currents and voltages, to regulate the power flow to the motor or end system. Additionally, all modern, high-performance motor drives require high-accuracy speed and position feedback information from the motor shaft. This information is required for a variety of different needs within the overall control structure. Relatively low-accuracy position information is needed for correct commutation of the motor currents in a three-phase machine; speed and position information, accurate to 8 to 10bits is generally sufficient for these tasks.

However, for closed-loop speed and position control, resolutions of up to 24bits on the rotor position are required for the most demanding applications.

To sustain their competitive advantages, developers and manufacturers of these high-performance systems are continuously increasing the precision, closed-loop controller bandwidths, and system reliability, while reducing overall system cost.

In UPS applications, the ability to achieve higher closed-loop bandwidths allows the generation of higher quality (less harmonic distortion) output voltages that are more immune to typical transient and distortion phenomenon found on the input ac mains. Similar criteria apply in servo systems in which overall system bandwidth is often a key selling point to help differentiate products from various vendors.

Advances in power semiconductor technology have enabled very high switching frequencies often in the tens of kilohertz &endash;to be achieved in the power converter topologies. As such, controller period (during which all of the mathematically intensive closed-loop control calculations must be performed) has been reduced to tens of microseconds. Today, typical servo drive update rates &endash; a critical measure of performance and often a key marketing differentiator between servo drives &endash; have been reduced to 50ms or less. It is generally accepted that traditional microcontrollers (MCU) lack the sheer computational capability to execute all of the necessary control functions in the allowed time.

The availability of on-chip hardware multipliers and architectures optimized for intensive mathematical calculation has propelled the adoption of DSP technology for these applications. Historically, these high-end, embedded control systems, have been primarily defined by their performance criteria and often times cost was a secondary concern.

Discrete solutions

Many manufacturers have developed discrete solutions to implement the control function of these embedded control applications. The total solution often comprises the DSP core and associated program and data memory, high-resolution (today 12bit is the accepted norm for these applications, with more demanding applications moving to 14bit), simultaneous-sampling, analog-to-digital conversion (for measurement of the feedback signals) and a variety of peripheral functions (often implemented with an FPGA device) for the generation of pulse-width modulation (PWM) signals to drive the power converter and interfaces to various rotor-position sensors.

Many critical application and technology factors must be considered when developing an integrated solution that does not compromise the overall system performance. For example, the integration of digital peripheral functions (such as PWM generation, encoder interface units, and serial communications ports) with the DSP core presents no significant technology challenges, since both are well adapted to standard digital CMOS processes.

A thorough applications knowledge is required in the specification of these peripheral functions if they are going to meet the needs of the end users. For example, a basic incremental encoder interface unit can be developed as a simple quadrature up/down counter. Unfortunately, this basic functionality does not provide any of the more advanced features that are required in real-world applications, such as flexible data latching of zero marker events, programmable filtering of encoder input signals for high-noise immunity and the accurate measurement of the time between encoder events for precise speed calculation. If sufficient care is not taken in the specification of these advanced features of the integrated peripheral functions, the user may be forced to add external logic (such as an FPGA device) to obtain the required performance. Clearly the overall benefits of the purported integration are lost in this situation.

Integrated solutions

Besides cost savings, there are numerous advantages to a fully integrated solution. In particular, a single-chip solution that combines the required mixed-signal integration simplifies the overall ease of use and cost of ownership for the customer &endash;it is easier to source a single device from a single supplier and integrate the device into a control board than it is to deal with multiple vendors/suppliers.

For many applications, developers of embedded control applications are using over-sampling and estimation techniques to extend the overall precision of the systems beyond the seemingly physical limits of ADC bits and DSP core bit width.

For example, a software implementation of a rotor position feedback, say a resolver to digital conversion, can achieve 16 to 18bit performance by acquiring the data using 12bit resolution ADCs and applying over-sampling methodologies. These techniques require high data-sampling rates; often the design challenge is in maintaining a high data throughput to the DSP core for processing. One approach is to integrate the ADC core and use direct memory access (DMA) channels to continually feed data from the ADC to the memory of the DSP core, balancing the computational capabilities of the core with the sampling rate of the ADC.

In addition, a single-chip solution keeps all of the high-frequency data lines from the ADC to the DSP on-chip, thereby reducing overall system-power consumption and easing EMI issues with the system design.

The integration of the mixed-signal components of the overall system&endash; analog-to-digital conversion, precision voltage references, power-on-reset circuits, on-chip amplifiers, etc. &endash; with the DSP core presents many interesting technology and application challenges. In many respects, the desire for integration of powerful DSP cores with high-sampling rate, high-resolution ADCs presents a host of conflicting technological issues.

Software development

On the one hand, developers of embedded control solutions recognize that one of the largest aspects of the overall project development budget is the software development. As such, the choice of a particular processor architecture for an embedded control design may depend on the quality and availability of good hardware and software development tools as much as the actual features of the particular processor device.

Software developers want to write the majority (if not all) of their controller code in higher-level languages such as C/C++. Often, very time critical tasks may still need to be optimized, either by careful use of the C language construction, the insertion of assembly language modules, or the use of library functions.

The ability to quickly identify the areas within the controller code that require optimizations and manage such optimizations offer a powerful tool for the targeted reduction of the overall software development effort. These software tools issues are pushing the performance requirements of the DSP core, such that ever higher instruction rates (or MIPS, millions of instructions per second), larger amounts of on-chip memory, larger addressable amounts of off-chip memory and special architectural features of the core to accommodate programming in C/C++ are required.

By definition, these requirements are pushing the use of the latest in DSP core technology built on low geometry CMOS processes, which operate at lower voltages and enable high-frequency capabilities (to hundreds of MHz today).

On the other hand, the desire for very high-performance ADC technology, with resolutions up to 14bits and fast sample rates (up to many MSPS, million samples per second) is made increasingly difficult as process geometry shrinks. With reduced operating voltages, and subsequent lower signal voltage swings, the ability to maintain the required signal-to-noise ratio (SNR) becomes increasingly challenging.

In addition, the close proximity of high-frequency clock rates in the digital design presents its own set of design challenges in effectively isolating the precision analog circuits from the noisy digital circuits and ensuring that there is no correlation or cross-talk from the digital domain to the analog domain.

Analog Devices has been exploiting its competencies in DSP and high-performance analog to create single-chip, embedded controller solutions that provide very high resolution ADC performance.

In particular, the ADMC300 integrates 5 channels of 12bit, SD (sigma-delta) ADC with a 25 MIPS DSP core. The ADMC401, introduced in 1999, provides 8-channel, 12bit ADC performance at very high sample rates of 6MSPS, with the same DSP core. Both of these devices have gained acceptance for many of today's embedded control applications.

Mixed-signal family

To meet the future needs of embedded control applications, and to target a variety of emerging applications, ADI has introduced a family of mixed-signal DSPs, called the ADSP-219xx. The first two members of this family are the ADSP-21990 and the ADSP-21991. These products combine an ADSP-219x DSP core, multi-channel, high-resolution ADC, and a selection of embedded control peripherals. As such, these products can be viewed as the merging of various technologies and competencies from with ADI &endash; the latest DSP cores (ADSP-219x) with the highest performance ADCs and embedded control intellectual property contained in our peripheral designs (leveraged and enhanced from the ADMC family of devices)&endash; to form a new class of mixed-signal DSPs.

Mixed signal DSP with embedded control peripherals

In particular, the ADSP-21990 offers 150MIPS, 16bit, 219x DSP core, 4K words of on-chip program memory, 4K words of on-chip data memory, and an external addressable memory space of 1Mwords (20 address lines). The mixed-signal integration consists of an 8channel, 14bit, 18.75MSPS ADC core (with dual sample-and-hold amplifiers for simultaneous sampling needs), precision on-chip voltage reference, and an integrated power-on-reset (POR) circuit.

The embedded control peripherals include a three-phase PWM generation unit (for control of power switching converters), 32bit incremental encoder interface (with noise filtering, flexible speed measurement hardware, a flexible reset, and latching modes), dual auxiliary PWM outputs (for control of secondary switching circuits such as power factor corrected front-end converters or dynamic dc link braking circuits) and a watchdog timer.

General-purpose peripherals are also provided and include three 32bit, general-purpose timers, a 16bit general-purpose I/O port, a memory DMA controller, and a flexible, peripheral interrupt controller. Two high-speed communications ports are included a synchronous serial port (SPORT) and a standard SPI port.

The devices are supported by ADI's CROSSCORE development tools. In particular, the development tools suite consists of the VisualDSP++ integrated software development environment that provides an assembler, C/C++ compiler, VisualDSP++ Kernel (VDK), advanced plotting tools, and statistical profiling to help quickly identify programming bottlenecks and reduce development time.

Hardware development tools consist of a low-cost EZ-KIT Lite evaluation platform. Functionality of the EZ-KIT Lite can be extended by the addition of JTAG in-circuit emulation (ICE) that permits full control for software debugging. Both PCI (Summit-ICE) and USB (Apex-ICE) versions of the emulator are supported.