S. Brigati, F. Francesconi, G. Grassi, D. Lissoni, P

0 downloads 0 Views 354KB Size Report
“An 0.8 µm CMOS Mixed Analog-Digital Integrated Audiometric System”. IEEE Journal of .... programmable division factor (PDF) is provided to the system.
S. Brigati, F. Francesconi, G. Grassi, D. Lissoni, P. Malcovati, F. Maloberti, A. Nobile and M. Poletti

“An 0.8 µm CMOS Mixed Analog-Digital Integrated Audiometric System”

IEEE Journal of Solid-State Circuits, 34, pp. 1160-1166, 1999.

1160

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

An 0.8- m CMOS Mixed Analog–Digital Integrated Audiometric System S. Brigati, F. Francesconi, G. Grassi, D. Lissoni, A. Nobile, P. Malcovati, F. Maloberti, and M. Poletti

Abstract—This paper presents a dual-channel fully integrated audiometric system, which generates the complete set of audio and control signals required for exhaustive audiometric tests. The system includes a novel signal generator, based on the direct digital synthesis technique, which fulfills the requirements of advanced audiometric tests. The proposed system faces two different problems, namely, the generation of a finely tunable pure sinewave and the generation of noise signals with a controlled spectrum. To achieve tuning capabilities down to 1 Hz at 20 kHz and 15 Hz at 100 Hz, a fractional division of a 40-MHz master clock based on noise-shaping techniques is performed. Moreover, for noise generation, a novel circuit based on pseudorandom sequences combined with analog switched-capacitor filters is used. The chip is fabricated in a 0.8-m CMOS process and occupies a 24.2-mm2 silicon area. It consumes 45 mW from a single 5-V power supply and achieves less than 090-dB crosstalk between the channels. Index Terms—Audio systems, mixed analog–digital integrated circuits, signal synthesis.

I. INTRODUCTION

M

ODERN medical instruments used for audiometric tests (audiometers) require high-quality audio signals and sophisticated signal processing. During audiometric tests, indeed, the human ear is stimulated with complex modulated sinusoidal signals or masking signals (noise) in order to identify hearing defects or losses, which can be corrected with hearing aids. The evolution of hearing-aid devices from simple amplification stages to sophisticated equalizers requires a substantial improvement in the quality of the audiometric tests to accurately customize the frequency response and optimize the hearing defect correction [1]. Therefore, the audiometers used so far based on analog components are no longer sufficient in terms of accuracy and precision [2]. This paper presents a mixed analog–digital integrated audiometric system (IASY), which generates the complete set of audio and control signals required for an exhaustive audiometric test. The proposed solution exploits direct digital synthesis (DDS) techniques to generate high-quality audio signals (pure tones and white noise) without the need for accurate analog components. II. SYSTEM IMPLEMENTATION

A list of the signals required for audiometric tests, as well as the specifications of a modern audiometer, are reported in Manuscript received October 16, 1998; revised March 24, 1999. S. Brigati and F. Francesconi are with Micronova Sistemi S.r.l., Trivolzio (PV) 27020 Italy. G. Grassi, D. Lissoni, and A. Nobile are with Amplifon S.p.A., Milano 20141 Italy. P. Malcovati, F. Maloberti, and M. Poletti are with the Integrated Microsystems Laboratory, University of Pavia, Pavia 27100 Italy. Publisher Item Identifier S 0018-9200(99)06171-5.

Table I. In view of the complexity of the signals, audiometric systems implemented with discrete analog and digital components are affected by crosstalk, electromagnetic interferences, and reliability problems. Moreover, the performance required can be obtained only by using well-matched components, fine tuning, and massive shielding of critical devices. A convenient solution to these problems is represented by the fully integrated audiometric system presented in this paper. The block diagram of the IASY chip is shown in Fig. 1 [3]. The core of the system is the digital signal generator, which consists of a variable-frequency, pure-tone (sinewave) generator and a white-noise generator [4]. The two digital words, representing the pure tone and the white noise, produced by the signal generator are converted into the analog domain by two 10-bit resistive stringwith an amplitude of 2 based digital-to-analog converters (DAC’s) and delivered to the analog filter section. Here, the white noise is filtered using suitable switched-capacitor (SC) filters to provide three additional noise signals with shaped spectra, namely, the pink noise, the speech noise, and the narrow-band noise. Last, either the pure tone or the noise signals, after reconstruction, are delivered to the output channels. The left and right output channels receive at their input all the generated signals (pure tone and noises) and, in addition, three external signals (microphone, tape, and compact disc), which allow the stimulation of the ear with prerecorded patterns. The left and right output signals are independently selected with analog multiplexers, according to the programming provided through the microprocessor interface. A programmable gain amplifier (PGA) is used to adjust the An 8-bit resistive stringlevel of the external signals to 2 based successive approximation analog-to-digital converter (ADC) allows the operator of the audiometer to monitor the signal level of the channels (VU meter). The selected output are delivered to two signals with a fixed amplitude of 2 external electronic attenuators and then to the earphones. Two 7-bit resistive string-based DAC’s in each output channel generate suitable analog control signals (RA and RS) to drive the external electronic attenuators. The digital words at the input of the DAC’s, representing the instantaneous signal level, are generated from the digital section or directly provided through the microprocessor interface, thus allowing complex modulations of the signal amplitude to be performed automatically. In view of the extremely large dynamic range of the human ear, a crosstalk between the two channels as low as 70 dB is required. The most significant functional blocks of the system, namely, the digital signal generator and the analog section (filters and PGA), will be considered in detail in the following sections.

0018–9200/99$10.00  1999 IEEE

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

LIST

OF

Fig. 1. Block diagram of the IASY chip.

AUDIOMETRIC SIGNALS

TABLE I SPECIFICATIONS

AND

1161

OF A

MODERN AUDIOMETER

1162

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

Fig. 2. Block diagram of the programmable divider.

III. DIGITAL SIGNAL GENERATOR A. Pure-Tone Generator The proposed DDS system [6]–[8] consists of a programmable frequency divider clocked at 40 MHz. The master is divided by a suitable factor to produce clock frequency as shown in Fig. 1. The the pure tone sampling frequency programmable division factor (PDF) is provided to the system by an external microprocessor. The clock signal obtained (CKL) drives a 6-bit counter, 10whose output word represents the address for the 6 bit lookup table (sine ROM) containing a sampled sinewave samples/period). The digital output code of period ( the lookup table is finally converted into the analog domain can be by a 10-bit DAC. The output sinewave frequency expressed as Fig. 3. Sigma–delta modulation effect.

PDF

(1)

is an additional division factor required to generate where CKF for the switchedthe clock signal at frequency capacitor filters used after the DAC. Assuming PDF to be an integer, the frequency resolution of the sinusoidal signal is limited. In worst case condiMHz, , and kHz), tions Hz, which is three orders of indeed, we obtain magnitude larger than the specifications. This limitation was overcome by using noninteger division of the master clock period. The digital word PDF actually consists of two coefficients: PDI and PDD, which represent the integer and fractional parts of the division factor, respectively PDI PDD). The word length of PDI is (PDF

determined by the minimum sinewave frequency (100 Hz), while the word length of the fractional part PDD is determined by the required resolution at the maximum sinewave frequency (20 kHz). Using (1), we obtain (2) Hz and (3) kHz Therefore, the required word length of PDF is 23 bits. Achieving fractional division of the master clock period while keeping spurious signals produced by distortion and

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

1163

Fig. 4. Spectra of the noise signals generated from the white noise.

phase noise below 60 dB is not straightforward. Therefore, to satisfy these requirements, oversampling and noise-shaping techniques were used too. The detailed block diagram of the programmable divider is shown in Fig. 2. The master clock bit synchronous signal MCK is accumulated by an counter (Counter). The resulting signal (CNT) is compared bit reference during each master clock period with an word (REF). When the two words become equal, the output signal of the divider (CKL) is set to one and the counter is reset. The fractional component of PDF (PDD) is applied to modulator, the input of a first-order digital sigma–delta whose output bitstream is added to PDI. The actual division factor is, therefore, alternatively PDI or PDI 1, depending modulator output (zero or one), thus on the value of the producing a nonuniform sampling of the output sinewave at as shown in Fig. 3. The noise-shaping effect of frequency modulator ensures that the distribution of zeros and the ones is sufficiently random. Therefore, the average frequency of the resulting sinewave, considering a suitable number of master clock periods (at least 2 ), is very close to the desired value, while the harmonic distortion and the phase noise introduced by the nonuniform sampling are negligible. By using the described technique, pure-tone frequencies in the range from 100 Hz to 20 kHz with a maximum resolution of 15 Hz at 100 Hz and 1 Hz at 20 kHz can be generated.

TABLE II POLE AND ZERO SEQUENCE USED TO REALIZE THE PINK-NOISE FILTER

A digital accumulator, clocked at 1 kHz, increments at each time step (1 ms) the value added to PDF, producing the required frequency shift, as shown in Fig. 2. The short increment sensitivity index (SISI) and difference limens for intensity (DLI) modulations are two different kinds of amplitude modulation performed on the pure-tone signal through the external electronic attenuators to identify particular pathologies. These modulations have been implemented in the digital domain. The two 7-bit words representing the signal amplitude (RA) and the modulation amplitude (RS) are converted into voltages by two 7-bit DAC’s and used to control the external electronic attenuator. C. White-Noise Generator

B. Frequency and Amplitude Modulations In audiometric applications, when hearing tests are performed in open spaces (without earphones), it is necessary to periodically modulate the frequency of the generated sinewave to 0.9 and back in 200 ms with 200 steps) (from to avoid the generation of stationary waves. This function, called warble, can be easily implemented with the proposed DDS technique. A shift register determines the frequency step according to

PDF

(4)

Noise sources are the second family of audio signals required in audiometric tests. Several noise signals, such as white, speech, pink, and narrow-band noise used as masking sounds, can be derived by properly processing a high-quality white-noise source. One of the most critical issues in the design of noise generators is, therefore, the choice of the primary whitenoise source. Analog instruments are typically based on high value resistors, which are very sensitive to interferences and require on-line calibration of temperature effects. By contrast, digital synthesis techniques based on pseudorandom sequence generators allow a perfectly white and repeatable spectrum to be obtained with very simple hardware [9].

1164

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

Fig. 5. Schematic of the pink-noise filter.

Fig. 6. Schematic of the programmable gain amplifier.

Fig. 8. Spectrum of a 2-kHz pure tone.

Fig. 7. Chip micrograph.

For effective audiometric tests, a repetition period of the pseudorandom sequence of at least 5 s is required. Therefore, assuming use of a clock frequency of 32 kHz, the random 10 samples long. Such a sequence has to be at least 1.6 maximal-length sequence can be generated implementing the eighteenth-order primitive polynomial (5) is then digitally processed The 19-bit output word to reduce the total word length to 10 bits, as required by the available DAC. However, simulations show that a simple truncation of the output word produces a low-pass shaping of the white spectrum. To overcome this problem, we introduced a compensating high-pass filtering, obtained by combining the

Fig. 9. Spectrum of a 2-kHz pure tone with warble.

19 bits of the pseudorandom sequence according to OUT

(6)

where OUT denotes the 10-bit output word. Basically, by combining the least significant bits with the most significant

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

1165

TABLE III DEVICE CHARACTERISTICS AND PERFORMANCE SUMMARY

bits delayed, we achieve a scrambling effect, which flattens the spectrum. This empirical correction has been optimized and verified with numerical simulations. IV. ANALOG SECTION A. Pure-Tone and White-Noise Filters After digital-to-analog (D/A) conversion, the pure-tone signal is bandpass filtered in order to remove any spurious tones superimposed on the generated sinewave. This task is performed by means of two cascaded conventional singleended SC E-type bandpass biquadratic cells centered at and sampled at 256 (the maximum sampling frequency is [10]. 5.12 MHz), with Likewise, after D/A conversion, the white noise is shaped with appropriate SC filters to generate the speech noise, pink noise, and narrow-band noise, whose spectra are illustrated in Fig. 4. The speech-noise filter is realized using a conventional single-ended SC low-pass biquadratic cell with 1-kHz cutoff frequency and 80-kHz sampling frequency in order to emulate the speech frequency content. The nonconventional (halforder) pink-noise filter with 100-Hz cutoff frequency and 10-dB/decade rolloff is obtained by cascading a biquadratic cell and a first-order cell, which implement the pole and zero sequence reported in Table II. This filter, whose schematic is shown in Fig. 5, implemented with a single-ended architecture and sampled at 80 kHz, produces a noise signal with constant power per octave. Last, the narrow-band noise is generated from the pink noise with a filter consisting of two cascaded

single-ended SC E-type bandpass biquadratic cells centered at and sampled at 256 (the maximum sampling frequency The resulting noise signal, is 5.12 MHz), with characterized by constant power independently of the selected is used as masking sound during pure-tone tests. value of B. Reconstruction Filters Since both signal generation and filtering are performed with sampled data techniques, reconstruction filters are required at the end of the processing chain to remove the in-band signal images and the high-frequency spurious tones. The reconstruction filters for either the pure tone or the noise signals consist of a cascade of SC and continuous-time (CT) cells. In particular, for the pure tone, we used an SC Etype low-pass biquadratic cell with 2- cutoff frequency and sampling frequency, followed by a CT Sallen–Key 256low-pass biquadratic cell with 20-kHz cutoff frequency [11]. By contrast, for the noise signals, we used an SC E-type low-pass biquadratic cell with 20-kHz cutoff frequency and 2.56-MHz sampling frequency, followed by a CT Sallen–Key low-pass biquadratic cell with 20-kHz cutoff frequency. C. Programmable Gain Amplifier The schematic of the programmable gain amplifier, used in the left and right output channels to adjust the level of the external signals, is shown in Fig. 6. The circuit consists of a single-ended to fully differential converter, followed by and ) a transconductor with the input transistors ( operated in the triode region [5] and a current-to-voltage

1166

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 8, AUGUST 1999

the main device characteristics and performance. One of the most important performances to note is the channel crosstalk, which is defined as the ratio between the amplitude of the pure tone programmed on one channel and the amplitude of the spurious signal coupled on the adjacent idle channel. Thanks to a careful layout, the channel crosstalk was kept below 99 dB. VI. CONCLUSIONS

Fig. 10.

Frequency response of the pink-noise filter.

converter. The gain of the PGA is tuned by changing the transconductance of the stage with an external variable voltage (VA) applied to the drain of the transistors operated in the triode region. By varying the control voltage from zero to about 180 mV, the gain of the amplifier ranges from zero to two. V. EXPERIMENTAL RESULTS The IASY system has been integrated in a 0.8- m doublepoly, double-metal CMOS technology. The chip micrograph is shown in Fig. 7. The chip area including pads is 24.2 mm while the total power consumption is 45 mW. The maximum frequency error of the pure-tone generator (i.e., the difference between the programmed and the measured frequency) over the whole operating frequency range is less than 15 ppm, corresponding to 0.3 Hz. pure Fig. 8 shows the frequency spectrum of a 2-kHz, 2tone. The magnitude of the second harmonic is about 83 dBc, leading to a total harmonic distortion (THD) of about 80 dB, while the signal-to-noise ratio (SNR) is about 90 dB. These values are almost constant over the whole audio frequency range. The effect of the 10% warble (frequency) modulation on the spectrum of the same 2-kHz pure tone is illustrated in Fig. 9. Fig. 10 shows the frequency response of the pink-noise filter. It can be observed that the 10-dB/decade rolloff has been obtained with reasonable accuracy. Last, Table III summarizes

In this paper, we presented a fully integrated mixed analog–digital audiometric system. The complete set of signals required for exhaustive audiometric tests is generated using direct digital synthesis techniques. A novel DDS architecture based on fractional clock division with noise-shaping techniques is used to achieve 1-Hz frequency resolution at 20 kHz, while a pseudorandom sequence generator is used to produce a high-quality white-noise signal. Although high-speed digital circuits (40 MHz) are integrated on the same chip with D/A converters and analog filters, we achieved 99 dB of crosstalk between the output channels. REFERENCES [1] J. F. Duque-Carrillo, P. Malcovati, F. Maloberti, R. Per´ez-Aloe, A. H. Reyes, E. S´anchez-Sinencio, G. Torelli, and J. M. Valverde, “VERDI: An acoustically programmable and adjustable CMOS mixed-mode signal processor for hearing aid applications,” IEEE J. Solid-State Circuits, vol. SC-31, pp. 634–645, 1996. [2] Clinic Diagnostic and Research Audiometer A460, Amplaid, Ref. 53803243, 1992. [3] S. Brigati, F. Francesconi, G. Grassi, D. Lissoni, P. Malcovati, A. Nobile, M. Poletti, and F. Maloberti, “An 0.8 m CMOS mixed analogdigital integrated audiometric system,” in ISSCC ’98 Dig. Tech. Papers, San Francisco, CA, 1998, pp. 116–117. [4] S. Brigati, F. Francesconi, P. Malcovati, M. Poletti, and F. Maloberti, “Digital synthesis of analog signals for audiometric applications,” in Proc. ISCAS ’97, Hong Kong, 1997, pp. 2625–2628. [5] U. Gatti, F. Maloberti, G. Palmisano, and G. Torelli, “CMOS triodetransistor transconductor for high-frequency continuous-time filters,” Proc. Inst. Elect. Eng., vol. 141, pt. G, pp. 462–468, 1994. [6] H. T. Nicholas III and H. Samueli, “A 150-MHz, direct digital frequency synthesizer in 1.25-m CMOS with 90-dBc spurious performance,” IEEE J. Solid-State Circuits, vol. SC-26, pp. 1959–1969, 1991. [7] P. O’Leary and F. Maloberti, “A direct-digital synthesizer with improved spectral performance,” IEEE Trans. Commun., vol. 39, pp. 1046–1048, 1991. [8] M. H. Perrot, T. L. Tewksbury III, and C. G. Sodini, “A 27-mW CMOS fractional-N synthesizer using digital compensation for 2.5-MB/s GFSK modulation,” IEEE J. Solid-State Circuits, vol. SC-32, pp. 2048–2060, 1997. [9] V. N. Yarmolik and I. V. Kachan, Self-Testing VLSI Design. Amsterdam, The Netherlands: Elsevier, 1993. [10] D. Johns and K. Martin, Analog Integrated Circuits Design. New York: Wiley, 1997. [11] R. Schaumann, M. S. Ghausi, and K. R. Laker, Design of Analog Filters. Englewood Cliffs, NJ: Prentice-Hall, 1990.

0