Hear this

Hear this Hearing must be one of the most important of the senses, and DSPs are coming to the aid of the deaf. Richard Ball explains how  
  Swiss precision… Phonak’s behind-the-ear hearing aids use its VLIW, parallel processing DSP architecture. It manages around 130Mips at just 10MHz clock frequency. By the way, they’re not normally transparent.
Small is beautiful… Hearing instruments that fit completely in the ear canal are the smallest available. This poses some considerable problems for the system designers, not least of which is getting the required DSP algorithms at the right power consumption.
Fancy processors are all well and good, but what real benefit does a microprocessor powered toaster bring?
Digital signal processing (DSP) on the other hand is bringing tangible benefits in the form of digital hearing aids.
Traditionally analgoue systems, the digital revolution has not passed hearing instruments by, and DSP is allowing manufacturers to use a whole new range of algorithms.
“Since the first hearing instruments appeared on the market the complexity has grown by about 100,” says Gos Leenen, head of DSP and IC development at the newly merged Beltone/Philips Hearing Technologies.. “The most complex devices on the market have more than one million transistors.”
Companies shifting to the brave new digital world include the recently merged Beltone and Philips, GNReSound, Phonak and Siemens Hearing Instruments. DSPs are also coming from Mitel Semiconductor, Motorola, Infineon Technologies and Texas Instruments.
“We are in one of the few fields where chip area is a concern,” Leenen at Philips points out. Just to make a chip and package designed to fit snugly in the ear canal is a major task. To then make it last a whole week before changing the battery is another incredibly difficult challenge.
Most companies use a two or three chip configuration. Philips has a complex Asic for all signal processing and data conversion, while the second chip is an EEPROM.
Philips’ Asic holds everything except the non volatile memory, including A/D converter, DSP, microcontroller and their associated RAM and ROM, oscillators and a special class D output stage. “Then we have quite extensive power supply circuitry,” says Leenen.
The DSP is designed specifically for the hearing algorithms required. “We use very long instruction words (VLIW) to do up to ten things in parallel,” says Keenen. “We want to do things in parallel to keep the clock speed low.”
Therefore voltage can be reduced and hence power is lower. More parallelism leads to larger area, but the trade off in terms of overall power consumption is better this way.
Current drawn by the device depends on the algorithm loaded at the time, which can be changed by the user. The least complex algorithms consume about 0.6mA, the most complex around 0.85mA. The latter gives about five days battery life.
The custom DSP approach has also been used by Swiss firm Phonak for its latest digital hearing instruments.
“We have a very special tailored architecture,” says Christian Berg, R&D manager at Phonak. Parallel execution units and a VLIW architecture allow for low clock speed and low voltage.
“The highest clock frequency we have is 10MHz, but we still have hundreds of Mips,” Berg points out.
Like Philips, Phonak insists that a custom DSP is needed in such a power sensitive application. Voltage and hence clock frequency must be kept low, which means using a parallel processor.
A general purpose DSP would have an order of magnitude greater power consumption, Berg claims:”They would have to drain 10mA.” Infineon’ Carmel DSP and hearing
Infineon Technologies’ Carmel DSP is also being touted for use in hearing applications. Like the Philips and Phonak devices, it uses a VLIW approach with multiple execution units. Carmel can perform up to 15 basic operations in parallel.
At normal voltages of 2.5V, power consumption is claimed to be 180mW at 120MHz. Dropping the supply to 1V (reduces power by 80 per cent) and dropping the clock down to a few MHz would bring Carmel into the realm of the hearing aid DSP.This idea of using VLIW and multiple execution units is gaining popularity in the DSP and microprocessor world. Trading off a larger area for a reduced clock speed results in lower power consumption, a trade off many designers are happy to make.  
Phonak uses several further techniques to reduce power. “We try to optimise the system for minimum memory access,” Berg says. The DSP uses gated clocks and the multiple execution units.
The very long instruction words to the DSP can be up to 650 bits long.
The DSP and pulse width modulation controller for speakers are fabricated in a 0.25?m digital chip, while three A/D converters, FSK remote control receiver and power management are made in a 0.35?m mixed signal process. Along with a 64k EEPROM, the chips are stacked together in a single package.
“At 1V it used to be exotic, but not any more,” says Berg.
In order to reduce problems as much as possible, Phonak always tries to use well understood semiconductor processes. “We have a 0.35?m mixed signal and a 0.25?m digital process and the technology is absolutely superb,” Berg says.
Using unknown, exotic technology is not on the cards:”You’re bound to run into trouble.”
Texas Instruments (TI) has a roadmap for its ‘C5000 family of DSPs that leads to 0.9V devices next year. These, the firm says will be targeted at, among other things, hearing instruments.
“We see the hearing aid market as very demanding in terms of power,” says Gweltaz Toquet, TI’s manager for C5000 cores. “Next year we will be able to provide 30Mips at 0.9V.”
However, the use of Mips as a metric can be misleading, as these types of processor often have special instructions replacing several simple instructions.
“The C5000 has some very specific instuctions such as FIR that can reduce the number of instructions and therefore the power consumption versus general purpose DSPs,” Toque t points out.
The evolution of hearing instruments from analogue systems through hardwired digital to the latest programmable DSP powered models has brought software to the party.
There is an increasing importance placed upon software, Toquet says. Hearing instruments are being designed that allow a choice of algorithm, depending on the environment and these can be tailored to an individual’s hearing loss.
Toquet says that although its lowest power C5000 devices will not ready until next year, designers are using existing chips to test out algorithms and system designs.

Leave a Reply

Your email address will not be published. Required fields are marked *