A low-power CMOS transceiver uses TDD techniques within unlicensed ISM bands in support of compact, low-cost wireless headset designs.
Conservation of power is a critical design goal in many wireless products, including wireless headsets for digital radio communications. One "power-wise" approach to wireless audio transmissions involves sampling, digitizing, and compressing audio signals before they are transmitted as the modulated portion of a RF carrier. Upon reception, these signals are decompressed and converted back to analog audio signals. By using time-division-duplex (TDD) techniques, it is possible to realize a low-cost, half-duplex RF transceiver (the CC1000 from Chipcon) that can transmit and receive audio signals with a low-component-count design that is suitable for wireless headsets.
Stricter regulations regarding the vehicular use of mobile telephones, as well as fear of RF radiation close to the brain, have prompted increased sales of wireless headsets for use with mobile telephones. Bluetooth has been touted as a "driving" standard for such applications as wireless headsets, but it remains to be seen if manufacturers of Bluetooth semiconductors can meet demanding market requirements for a low price. As a result, designers of wireless headsets are anxiously seeking low-cost, low-power alternatives to Bluetooth integrated circuits (ICs), such as the single-chip model CC1000 RF transceiver from Chipcon (Oslo, Norway).
Based on 0.35-µm complementary-metal-oxide-semiconductor (CMOS) technology, the CC1000 represents a combination of low cost, high integration, high performance, and flexibility in a low-power device. Designed primarily for frequency-shift-keying (FSK) systems in the industrial-scientific-medical (ISM) bands at 315, 433, 868, and 915 MHz, the single-chip transceiver can be programmed for use from 300 to 1000 MHz. It operates on supply voltages from +2.1 to +3.6 VDC, and only consumes 7.4 mA at +3 VDC in receive mode. The leakage current is only 0.2 µA, ensuring miniscule draw on batteries during nonoperating times. The transceiver chip features receiver (Rx) sensitivity of −110 dBm (at 433 MHz and 1.2 kb/s) supports data rates from 0.6 to 76.8 kb/s.
In a digital telephone system, analog voice signals are converted to digital signals through pulse-code modulation (PCM), using three steps: sampling, quantizing, and coding. Band-limited (4-kHz-wide) voice signals are sampled at a rate of 8 kHz (in agreement with Nyquist theory). The amplitude of the voice signal is sampled (measured) every 125 µs. Each sample is then quantized (or truncated) into a number, usually an integer. For example, if 13 b is used, this integer could be between −4096 and +4095. Coding is the way these quantization levels are represented as digital numbers. For example, the coding method known as "two's complement" may be used to express negative, as well as positive, numbers. Most of the information captured for human speech has a small-signal character, contained within higher-resolution, smaller amplitudes. For less-likely larger-amplitude voice signals, the use of a uniform quantizer (equally spaced digitizing steps) provides high, but unnecessary, quality. The uniform quantizer can also yield pronounced truncation effects for the more frequent small-amplitude signals. As a result, using nonuniform quantization provides a system that is more appropriate for human speech.
Nonuniform quantization can be achieved by passing a voice signal through a compressor at the transmit end, and then passing the signal though an expander on the receiving end as part of a process known as "companding." By using companding, the required code-length word can be reduced from 13 b to 8 b or less, while retaining the subjective quality of the voice signal. Companding can be performed in hardware in a coder/decoder (codec) circuit or in software using a look-up table or a real-time calculation. Two international standards for creating 8-b encoded data are u-law and A-law. In the US and Japan the accepted standard is u-law, while A-law is used in Europe.1 The 8-kHz sample rate combined with companding leads to an 8-b code word. The digital voice stream is therefore said to be 64 kb/s.
A public-telephone network is an example of a full-duplex system, where speech is transmitted in both directions at the same time. In a half-duplex system, speech travels only in one direction at one time (similar to a walkie-talkie system). As an extension of the public-phone network, a wireless headset must, therefore, be full duplex. Full-duplex systems require more complex circuit solutions than half-duplex systems, where the Rx and transmitter (Tx) can share many of the system function blocks. A half-duplex approach can save space and cost.
By using TDD, full-duplex operation can be achieved with half-duplex cost and simplicity. Using TDD, the signal is transmitted one way at a time, but the direction of transmission is switched very fast with a small latency (time delay). As long as the latency is in the order of 100 ms or less, the human ear will not detect it and a normal conversation can occur. In comparison, satellite telephone systems have a large delay, making a normal conversation flow more difficult. Using TDD makes it necessary to buffer the digital voice stream from one party while the other is transmitting. A voice stream of 64 kb/s, therefore, would require a wireless TDD data link of at least 128 kb/s. The turn-around time from Rx to Tx would require an even higher data rate.
High data rates require larger RF bandwidths or advanced modulation techniques. To reduce the digital-voice data rate to be transferred, another layer of coding is required, such as Adaptive Differential Pulse Code Modulation (ADPCM) or Continuous Variable Slope Delta coding (CVSD). Each scheme is based on differential coding—instead of sending the absolute value of the sample, the difference between the current sample and the previous one is sent. Using ADPCM coding, 8 b are coded down to 4, 3, or 2 b. A data rate of 64 kb/s is then reduced to 32, 24, or 16 kb/s. Assume now the use of a 32-kb/s ADPCM, since this form of compression does not place a heavy burden on the system microcontroller. Dedicated hardware implementations could also be used for the compander and the ADPCM codec.
Figure 1 shows the block diagram for a wireless headset solution, with the signal flow shown in Fig. 2. A functional diagram of the CC1000 is shown in Fig. 3. As can be seen, the IC contains components for the Tx, Rx, frequency synthesizer, and for control functions. Few external components are required for operation, allowing the transceiver IC to support applications requiring low-target selling prices, such as wireless headsets. At 433 MHz, the CC1000 consumes only 7.4-mA current when receiving and only 10.4-mA current when transmitting a 0-dBm signal.
With TDD, it is important that the Rx/Tx turn time is short to keep the overhead down and, therefore, the overall data rate down. The CC1000 phase-locked loop (PLL) provides turn-around times of less than 200 µs. This is also important in the idle mode where the earpiece is polling for an incoming call. When polling for an incoming call, the received-signal-strength-indicator (RSSI) circuitry can be used to quickly determine whether or not there is a signal. The CC1000 provides an analog RSSI signal that can be connected to an analog-to-digital converter (ADC), preferably integrated in the microcontroller unit (MCU). The CC1000 contains a programmable PLL that support the use of several frequencies or channels. If one channel is occupied, an automatic channel-selection algorithm can be used to find a free channel to avoid interference with other radio systems in the same frequency band. The integrated bit-synchronizer in the CC1000 internally samples, filters, and slices the received data, easing the burden of the MCU. The received data-stream (DIO) is provided together with a synchronous clock (DCLK). Typically, an interrupt pin of the MCU is used for the clock.
In TDD (Fig. 4), the time during which each side transmits is known as the dwell time, while the guard time is the time needed to change the direction of transmission. Overhead can be reduced by keeping the dwell time long and the guard time short, but using a long dwell time introduces longer latency. A dwell time of 50 ms represents a good compromise, providing a repetition rate of 100 ms or 10 Hz. A first-in, first-out (FIFO) buffer is used to buffer the data sent and data received during TDD. The FIFO is usually implemented in the microcontroller random-access memory (RAM) as a software FIFO. A state machine in the microcontroller controls the TDD and adds the control information. The 32-kb/s voice data is therefore transmitted at 76.8 kb/s. This includes the minimum bidirectional voice-data requirement of 64 kb/s, additional control information, and guard time.
Figure 5 shows a suggested packet format. The frame consists of a preamble that is used by the Rx to synchronize to the incoming data and set the slicing threshold in the data slicer. The "start of frame" is a unique term used to separate the preamble from the address and control information field. The bulk of the packet will be the voice data. A check sum should follow the voice data and can be used by the Rx to check the integrity of the received data.
A quality-of-service (QoS) measure can be defined based on the received data. A Cyclic Redundancy Checksum (CRC) can be added to provide a simple QoS solution. The QoS measurement can also be used to make an automatic frequency selection if the present channel is not working satisfactorily. With a digital radio system, scrambling or even encryption can be added to protect the privacy of a transmission. In most cases, scrambling is sufficient because the range of the system is small compared to what can be eavesdropped acoustically. Directional antennas could be used in a potential eavesdropping scenario, but very sensitive or confidential information should not be shared over a telephone in any case. Also, it would be pointless to make a wireless headset more secure than the rest of the telephone network. Encryption is a less attractive option because it significantly increases signal-processing requirements when implemented in software, defeating the benefits of a low-cost, low-power solution.
In idle mode, a wireless earpiece waits for an incoming call. This is performed using Rx polling which means that the Rx is turned on at regular intervals in search of a valid signal (Fig. 6). A typical polling interval is 1 s: once every second, the wireless earpiece Rx is turned on and looks for a signal. First, the RSSI can be checked. If a signal is detected, the MCU starts to look for the preamble in the received data stream. It requires 11 b to fill the averaging filter in the data slicer and from 4 to 8 b to detect a valid preamble. Thus, a single poll requires a minimum of 19 b-periods or 250 µs using 76.8 kb/s. In addition, it takes approximately 250 µs for the PLL to lock after crystal startup. In the worst case, each poll is then 0.5 ms, yielding a polling ratio of 1 s / 0.5 ms = 2000:1. The contribution to the average current consumption in idle mode is, therefore, less than 10 mA/2000 = 5 uA.
The ADC/DAC, compander, and ADPCM or CSVD can be implemented in various ways (Fig. 2). Some manufacturers provide hardware solutions integrating several or all these functions, but digital-signal-processing (DSP)/MCU solutions can also be used. For example, models CMX639 (CVSD) and CMX649 (ADM) are two voice codecs provided by CML Microcircuits (www.cmlmicro.com), while models MC145540 and MC145481 are ADPCM codecs from Motorola (www.motorola.com). Several ADPCM codecs, including compander functions in the ML7029 and MSM7540L/7560L/7570L/7590L ICs are available from Oki Semiconductor (www.okisemi.com). ADPCM coding and u-law/A-law companding can also be performed in an MCU (from Cypress Semiconductor) or DSP (with solutions available from suppliers such as Texas Instruments and Lucent Technologies). Finally, the ADSP-21ESP202 series of embedded speech processors from Analog Devices (www.analogdevices.com) combine the ADSP-218x 16-b fixed-point DSP core, advanced mixed-signal technology, and software intellectual property (IP) for high-performance speech-processing applications.
In the US, the use of radio Txs and Rxs is regulated and approved by the Federal Communications Commission (FCC). A few frequency bands, the so-called ISM bands are dedicated to license-free operation. The most attractive of these bands is 902 to 928 MHz as regulated by FCC CFR 47, part 15.249.5 The Tx is allowed to generate a field strength up to 50 mV/m as measured at 3 m, equivalent to −1-dBm effective radiated power (ERP). This is more than sufficient for the very short range required in this application. The frequency band is 26 MHz wide, which makes it possible to use several channels at the same time. For example, the channels can be placed at 1-MHz spacing, supporting up to 24 channels.
In Europe, two unlicensed radio bands below 1 GHz should be considered: 433 MHz and 863 MHz. The latest revision of the CEPT 70-03E recommendation3 notes that "audio and voice signals should be avoided in the band 433.05 to 434.79 MHz." This implies that national bodies in Europe could totally ban such applications in this band (which is not yet harmonized). Therefore, this frequency band should not be used for wireless headsets targeting the EU/EEA countries. A better option is the band at 863 to 865 MHz that are specifically set aside for wireless headset applications. Up to 10 mW (ERP) is permitted in this frequency band. The minimum recommended channel spacing when the CC1000 is used is 500 kHz, supporting three or four channels in the 863-to-865-MHz band. Consult ref. 2 for a summary of regulatory issues.
A multichannel system could be an advantage in a crowded area to avoid interference from other wireless headsets or systems using the same frequency band. A software algorithm could select the channel or frequency automatically. For example, the base at the handset could use the RSSI to scan for a free channel and choose the best at any moment when setting up a new session (a new incoming or outgoing call). The earpiece could then simply scan all channels when polling in the idle mode. This scheme would increase the power consumption in the idle mode of the earpiece due to the scanning, but it would also improve the reliability of the system.
The co-location properties of a radio system are usually expressed in co-channel and adjacent-channel rejection. The co-channel rejection is the strength of the interfering signal compared to the wanted signal when operating at the same channel. Due to the properties of the frequency-modulated (FM) demodulator, the weakest signal will be suppressed in the demodulation process. Using CC1000, the interfering signal can be up to −3 dBc, which is 3 dB below the wanted signal.
Blocking performance is a measure of the ability to withstand strong out-of-band signals. As the cell phone itself is operating at a few megahertz away from the operating frequency of the system (863 or 915 MHz), it is important to provide sufficient out-of-band selectivity. For a frequency offset of ±1 MHz, the CC1000 delivers 43-dB blocking (compared to 30 dB required by the EN 300 220, Class 2 standard4). For an offset of ±2 MHz, the CC1000 offers 49-dB blocking (compared to 35 dB required by the EN 300 220 standard). For an offset of ±3 MHz, the CC1000 achieves 68-dB blocking (compared to 50 dB for EN 300 220). For an offset of ±4 MHz, the CC1000 reaches 72-dB blocking (compared to 60 dB for EN 300 220).
Power-supply design in a headset is very important for the total battery lifetime. The use pattern of a wireless headset implies a rechargeable battery technology. This could be nickel metal hydride (NiMH) or lithium ion (Li-ion). One- or two-cell boost converters are available from several vendors, such as the TPS61000 series from Texas Instruments.
REFERENCES
