For the past 20 years, clinical physiologists have been writing papers on how noise from airplanes, trains, automobiles, air conditioning, television sets, stereo systems, vacuum cleaners, toys, and computers negatively affects the health, well being, short and long term memory, cognitive abilities, motivation, and attention span of adults and children who are subjected to large and small amounts of noise. This large amount of research leads to one compelling conclusion, "Quieter is better."

Nowhere has this work been more evident than in the classroom. To ensure that noise levels in schools are at acceptable levels, in 1999 the Access Board, which develops accessibility standards under the Americans with Disabilities Act (ADA), voted to collaborate with an existing Acoustical Society of America (ASA)/ANSI Working Group on Classroom Acoustics to develop recommendations for classroom acoustics. The Group completed a final draft in January 2001 and submitted it for review and ratification to the ASA/ANSI Committee on Noise. Soon, background noise standards of 35 dB will be enforced by building codes.

As more adults and children spend more and more time near personal computers (PCs), the ergonomic and environmental demands on these machines increases. Today there is an increasing awareness, both among PC users and manufacturers, that a low level of PC acoustic emissions must be regarded as an important factor in a comfortable working and home environment. End users are unwilling to accept yesterday's solutions of over design: putting more and more fans on a desktop computer, thus making it sound like a small airplane.

The Pentium takes its temperature

To ease the burden on cooling solutions, Intel has integrated a new thermal control circuit (TCC) feature into the Pentium 4 processor. The TCC limits the processor temperature by modulating (starting and stopping) the processor core clocks when the hot spot of the die is very near the limits. When the temperature has returned to a non-critical level, the TCC goes inactive and clock modulation ceases. This trip point thermal sensor, which ensures that processor temperature in the hot spot does not exceed safe parameters is, in essence, a safety mechanism. When the TCC is active, and clocks are being modulated, and a pin on the processor is asserted.

By taking advantage of the TCC, system designers and integrators may decrease the cooling system cost and reduce acoustic noise, while maintaining processor reliability and performance goals. Other options within the thermal-management logic allow end-user-controlled software to monitor and control the thermal and acoustic characteristics of the system.

It is important to understand the relationship between thermals and acoustics. As fans spin faster, the large amount of moving air creates noise, but the system is cooled by the maximum possible amount. As the fans slow down, acoustics are reduced, but the ability to remove heat from the system is also reduced, and the components get hotter. The optimal solution will be to move just enough air to ensure that the temperature limits of the system are met, while minimizing the acoustics, thereby running the system at the maximum safe temperature in all environments.

There are two main issues when dealing with processor thermals:

ensuring that the thermals do not exceed functional limits. This is function of the thermal monitor found on the Pentium 4 processor. tracking overall temperature. This is the function of the thermal diode, and the external thermal diode monitor.

This monitoring function is away from the hot spots and is less likely to see high-speed thermal transients. The thermal diodes can be used for automatic fan speed control, to keep the system as quiet as possible.

Figure 1a: Processor and heat sink temp showing a thermal event, while fan speed is at 100%.

Figure 1b: Processor and heat sink temp showing a thermal event, while fan is controlled by processor temperature.

The real test is to implement automatic fan control mode, and see if it does in fact, affect processor thermals. The combined effects of qJA (junction-to-ambient thermal resistance) can be seen in figure 1a, where processor thermal diode temperature and heat sink temperature are plotted vs. time. One can easily see that the temperature of the heat sink lags behind the processor temperature by a large amount, because of the large thermal mass of the heat sink. From this, one can extrapolate that if the fans were not spinning quite as fast, or if the heat sink temperature were slightly higher at the beginning of the thermal event, it would not make a difference in the overall performance of the system.

This is shown in figure 1b, where the heat sink fan speed is controlled via the on-chip processor thermal diode. As processor temperature increases, fan speed will increase; as processor temperature decreases, fan speed will decrease. As fan speed is decreased during the idle times before and after the thermal event, the processor and heat sink are running marginally hotter than if the fan was running at full speed. However, even at the worst case, the processor temperature does not exceed normal operating parameters. The rate-of-change of fan-speed is controlled so that negative psychoacoustic effects will not be created. The human ear is much more sensitive to changes in sound levels than in the actual sound level.

Figure 2: The duty cycle for the fan control motor rides the ambient temperature ramp of the Pentium.

Fan control revealed

The methodology of AFC is shown in figure 2. From a cool state, as temperature increases (red line), and approaches TMIN, the fan stays off. Once the temperature reaches TMIN, the fan will turn on to the minimum duty cycle. As the temperature continues to increase, the fan speed will also increase until it has reached 100% at TMIN + TRANGE. As temperature decreases (blue line), the fan speed will decrease until the temperature reaches TMIN. In order that the fan not continually cycle on and off, the fan will continue to run at the minimum speed until it reaches the hysteresis point. Here it will turn off, and wait until the temperature rises again.

Automatic fan control (AFC) does not adversely affect the thermal performance of the system, and processor temperature is the best possible input, so fan control based on processor temperature for acoustic betterment of the system can be explored.

Implementing continuous measurement and dynamic adaptation to temperature makes it possible to decrease the acoustic levels of PCs, and to lower their power consumption.

The temperature at which TCC trips can vary from processor to processor, and individual systems will have different heat sinks, different fans, different chassis air flows and different amounts of heat generated, so it is impossible to indicate the most appropriate TMIN and TRANGE for every system. Even with systems that have the same chassis and the same internal components, the variability of processor, assembly, altitude, and other parameters make pre-defining the optimal TMIN impossible. If the TMIN is too low, excess acoustics are generated. If the TMIN is too high, TCC clock modulation will occur, and performance will be impacted.

By using self-calibration, each system can tune its own acoustic response. The addition of an operating point allows TMIN to self adjust. If the actual temperature exceeds the operating point, then TMIN is reduced. As TMIN decreases, the slope of the control line is constant, so the fan speed increases at the same rate.

Figure 3: The temperature at which the fan turns on ( TMIN) is adjusted to minimize noise yet keep the thermal control circuit sensitive.

Figure 3 shows that TMIN has self adjusted to the new value, tMIN, in order to keep the actual temperature below the operating point set by TOPER.

The ADM1027 System Monitor and Fan Controller monitors the Pentium 4 processor's PROCHOT# pin, so it can self calibrate to just below the temperature at which the TCC just becomes active. In this way, the platforms will run at the hottest safe temperature, without running into thermal related performance issues. And the hotter the system, the quieter the system.

Figure 3: The temperature at which the fan turns on ( TMIN) is adjusted to minimize noise yet keep the thermal control circuit sensitive.

The self-calibration ability give the system independence from variations of fans, ambient temperature, chassis design, processor heat sink, thermal interface material (material between the processor and heat sink). Also discounted: altitude, networking cards, CD-ROMs, DVDs, hard drives, memory, chipsets, and other PCI cards, and computer placement-all creating or moving unknown amounts of heat in the system.

It is only with this type of self-calibrating functionality that the PC manufacture can produce a system that is truly quiet while maintaining optimum performance.

Making audio sound better

Audio subsystem design for PCs and other consumer electronics devices is typically based on an inverse relationship between the cost of the design and the quality of audio output achieved. Most designs today have to make tradeoffs that compromise affordability for high quality audio, or compromise quality for lower costs.

New AC'97 codecs provide the opportunity to change this relationship by providing 20-bit high performance DACs and DSP filtering features, along with the ability to easily set filtering parameters through the use of configuration software. The result is a greater set of design options available to engineers and the ability to take speaker alternatives into account when delivering a full system audio solution.

When the 20-bit DACs on Analog Devices' newest codecs are compared to $99 PCI Sound Cards, using industry standard audio tests on an Audio Precision System 2, the analog performance meets or exceeds that of the PCI sound card. The DAC performance of the AD1981A AC'97 codec also approaches the performance found in the same $99 PCI Sound Card.

While it is interesting to look at signal-to-noise ratio (SNR), total harmonic distortion plus noise (SINAD) and frequency roll off graphs, they do not describe how good the audio really sounds. Audio performance specifications describe what a voltage looks like on a pin, not how something sounds. The only instrument that is able to differentiate how something really sounds is the human ear. The most important aspect of how something sounds is the devices that takes the voltage and creates the sound waves that we hear: the amplifier and the speakers.

Making speakers sound better

In the classic no-frills engineering solution to audio reproduction, audio is typically played without equalization. Engineers cannot adapt the resulting output to a particular set of speakers or headphones, and the output cannot be adjusted by end users. Because the output can't be adapted to the audio reproduction hardware or to the acoustics of the surrounding environment, it is highly unlikely to be identified by the human ear as a true reproduction. Some frequencies will appear to be highlighted, while others will sound subdued. The audio will be reproduced, but not in its seemingly original form.

The Analog Devices AD198x family of codecs provides the opportunity to significantly change the audio quality that an end user hears by providing DSP bandpass filtering features as a part of the codec- along with the ability to easily set filtering parameters through the use of configuration software. The result is a greater set of design options available to engineers and the ability to take into account speaker alternatives in delivering a full system audio solution.

A key design goal for the AD198x family was to deliver a combination of a fixed AC'97-compatible codec with software configurable speaker EQ DSP features. The AD1981A offers significant design advantages over any of the alternatives available, notably software configurable speaker equalization. Digital circuitry built into the AD1981A enables the PC motherboard designer to adjust the center frequencies and bandwidth and to be able to tune for different speaker sets. And, because the DSP features are integrated with the codec, the manufacturing cost of the subsystem can be reduced without compromising on audio quality. This software configurable approach addresses the challenge of quality variability that can be introduced to the audio subsystem at manufacturing time, while also offering the potential to control costs by replacing discrete components with an integrated chip.

The AD1981A provides OEM audio engineers with the ability to reduce costs, improve audio quality and manufacturability, and increase the range of design options available. It offers significant improvement in cost by requiring fewer components to support a mid-to-high level of audio quality, while enabling better control of acoustic design and audio quality.

Overall, incorporating the AD1981A or AD1980 into an audio subsystem design affords engineers a better audio quality from a given set of speakers, or a similar level of quality from less expensive speakers. By doing so, it provides greater leeway over speaker selection as a part of the complete system solution.

Multi-Channel (5.1) Audio Output

With the introduction of high performance, multi-channel AC'97 codecs, motherboard designers who want to offer home-theater-like systems now have a solution, at an affordable price point. The AD1980 offers 6 DACs for front stereo, surround stereo, and center / LFE channels, and 2 ADCs for stereo capture of line and microphones for array processing.

With the popularity of DVD ever increasing, and with different digital surround formats being introduced, taking advantage of the host processor to do the digital decoding, and passing this to low-cost analog speakers, gives the PC OEM the flexibility to provide low-cost multi-channel solutions to their customers.

Mic processing for enhanced communications

With the advent and distribution of various H.232 video and voice codecs, more and more people are using the Internet to do voice phone calls. Nowhere in the audio path are things more susceptible to electrical noise than on the microphone input. Something needed to be done to clean up the microphone input, to reduce background hiss and electrical noise, and to make voice-over-IP applications sound better that AM radio, and speech recognition applications that actually work.

PureAudio 2.0 is a digital noise-cancellation algorithm designed to sample an ambient noise environment and to attenuate the noise sources around the desired speech signals, thus delivering a pure audio signal. As a result, continuous and repetitive noise is removed from the audio input. When used with speech-enabled applications, recognition system latency is markedly improved, and residual digital distortion is significantly decreased. Designed specifically to improve signal-to-noise ratio, PureAudio 2.0 works best in canceling stationary noises such as computer fans, motherboard electrical noise, air conditioning, etc.

PureAudio 2.0 enhances a wide variety of speech-enabled applications including desktop speech recognition, Internet telephony, videoconferencing, multi-player gaming, voice verification, voice chat, or wearable computers. In essence, PureAudio 2.0 can be applied to any audio-input application.

PureAudio 2.0 runs completely on the Intel Pentium 4 processor. In addition, the software can be utilized as an independent noise cancellation algorithm for single-element microphone solutions, or can be used to enhance other solutions including DSDA 2.0.

Figure 5: Directionality is the key to noise cancellation in voice recognition systems.

Array Mics for voice recognition and control

To make voice applications work, the end user cannot be tethered (have a cord from a headset) to his or her PC. Usability studies have shown that most end users would prefer to talk in a natural environment, and do not want to put on a headset.

DSDA 2.0 is a sophisticated and robust noise cancellation solution developed to bring a new level of clarity to voice communication applications. Using adaptive array microphone technology enables the optimal performance of headset-free, far-field voice input by creating a narrow reception cone of microphone sensitivity on the user's voice and canceling noise outside of that signal by using a low cost analog stereo array microphone.

DSDA version 2.0 utilizes a unique de-reverberation technique, which dramatically reduces reverberation noise caused when a speaker's voice echoes from walls or ceilings, which has the effect of degrading the performance of speech recognition applications. As a result, this software offers greater sensitivity and a superior solution for clear voice recognition with untethered, far-field voice communications.

There are many opportunities for PC motherboard manufactures to differentiate their products based on low cost functionality that end users are looking for, to be able to sell their PCs more effectively. Quieter PC Better Audio while playing games Enhanced voice communications Low cost voice recognition that works.