A digital preamp system described
A block diagram of a digital preamplifier and its interface is shown in Figure 2. The microphone element signal is first amplified, and then converted to digital by the A/D converter. These blocks receive their power from an internal regulated supply, ensuring good power supply rejection and independent digital and analog supplies.
The preamplifier is built using two operational transconductance amplifiers in an instrumentation amplifier configuration where the gain is set by matching of the capacitors. This configuration allows near infinite input impedance, which is desirable for capacitive signal sources, as the source sees a MOS transistor (MOST) gate.
Figure 2. In a digital microphone system, the microphone element signal is first amplified, and then converted to digital by the A/D converter.
(Click on image to enlarge)
The use of capacitors for gain setting allows high gain accuracy, limited only by process lithography, and the high linearity inherent to poly-poly capacitors. The gain of the amplifier is easily set by metal mask programming, allowing gains of up to 20 dB.
The A/D converter is a fourth-order, single-loop, single-bit sigma-delta modulator, whose digital output is a single-bit oversampled signal. Using a sigma-delta modulator for A/D conversion holds several advantages:
- High accuracy can be obtained without imposing severe matching requirements for the circuitry, due to the fact that noise shaping of the quantization noise in the frequency domain shifts the quantization noise upwards and out-of-band.
- The A/D converter is inherently linear as a single-bit sigma-delta modulator is used.
- Low power consumption, particularly for the single-bit, single-loop modulator, as only one integrator has severe design constraints imposed. All inner loop integrators have their outputs noise-shaped and thus have relaxed design requirements.
Sigma-delta modulators are prone to instability when the input exceeds a certain maximum stable amplitude (MSA). Higher-order modulators (>2) that go unstable due to overload never return to stable operation even when the input is reduced below the MSA. To counter potential instability, a digital control feedback system alters the sigma-delta noise transfer function, forcing the modulator back into stable operation.
A built-in power-down mode lowers current drawn by the system from 400 A to approximately 50 A. The power-down feature allows the user to conserve power whenever the microphone is not needed. The start-up time from power-down is 10 ms. Power-down mode is entered by letting the system input clock frequency fall below 1 kHz.
Three dominant noise sources in CMOS preamplifiers for capacitive microphones are flicker (1/f) noise, wideband white noise from the input transistors and low-pass filtered white noise from an input bias resistor. An input bias resistor RBIAS, is needed for setting the amplifier DC operating point. A-weighting is applied to emulate the human ear spectral noise shaping.
Flicker noise has an inverse dependency on transistor area and is input referred given by
V2(f) = Kf/(WLCoxf)
Where Kf is a process dependent constant, f is frequency, W is the MOS width, L its length and Cox is the gate capacitance per area. Hence we can control 1/f noise by choosing sufficiently large input transistors.
The input referred white noise is inversely proportional to the MOST transconductance gm:
V2(f) = 8/3kT/gm
For a MOST in strong inversion, the gm is approximately given by; gm=2Id/Veff, where Id is the drain current, and the effective voltage, Veff=Vgs-Vth, is the gate-to-source voltage minus the MOST threshold voltage Vth.
By designing the input pair to be very wide, a bipolar-like mode of operation is imposed upon the MOST as it enters the weak inversion operating mode. Here, gm is given by: gm=Id/(nVT), where n is the slope factor (typically 1.5) and VT is the thermal voltage. Thus, optimum white-noise performance is achieved by maximizing the MOST aspect ratio.
The input bias resistor is connected to a capacitive source, so its noise will be low-pass filtered. We assume that the noise is low-pass filtered white noise and the cut-off frequency is much smaller than the audio band frequencies. It can be shown that the total noise power is
where k is Boltzmann's constant, T is temperature in Kelvin and C is the capacitance connected to the node.
As the trend continues for smaller microphone cartridges with lower cartridge capacitance, this noise source will increase as the microphone cartridge capacitance decreases. The audio band noise power coming from the bias resistor, though, will also depend on the cut-off frequency of the low-pass filter.
The lower the cutoff frequency, the smaller the portion of the total noise power will be in the audio frequency range. So, in order to keep the noise low, the value of the bias resistor value will have to be increased as the microphone capacitance is lowered.
The impedance will have to be increased by a factor of four each time the capacitance is decreased by a factor two. For a 3 to 5-pF microphone capacitor, the resistor should have a minimum value of approximately 10 GΩ.
A good solution for implementing such a large value resistor on chip is a pair of cross-coupled diodes, which have a very large resistance around equilibrium-typically 1 to 10 TΩ. The resistance decreases for larger signals, assuring fast settling after overload situations. Figure 3 shows the in-band noise as a function of RBIAS.
Figure 3. The in-band noise of a CMOS capacitive microphone preamplifier as a function of RBIAS.
(Click on image to enlarge)
The area of the input transistors of the preamplifier must be optimized to the microphone capacitance. The 1/f noise from the input devices will decrease if the input devices are made very large, but the capacitive loading of the signal source will also increase. As a result the signal will be attenuated and the signal-to-noise ratio (SNR) reduced.
On the other hand, if the input device is made very small, the capacitive loading of the signal source becomes insignificant, but the 1/f noise increases dramatically. As a result, the SNR will also be poor.
An optimum point exists where the SNR is maximized. For 1/f noise, this optimum is found when the gate-to-source capacitance of the input device equals the microphone capacitance plus parasitic capacitance. A similar optimum exists for white noise, but here the optimum is found when the gate-to-source capacitance of the input device equals 1/3 of the microphone capacitance plus parasitics. In practice, the gate capacitance is chosen in between the two.
Bootstrapping minimizes the input pad contribution to the overall chip input capacitance. As the output-referred white noise is proportional to gm, all current source MOSTs are biased in the strong inversion region ensuring minimum noise contribution.
The table below shows the key figures and performance of the ADAU1301 microphone preamplifier.
(Click on image to enlarge)
This amplifier fulfills the needs of ECM elements well, but is not suitable for the emerging MEMS microphone elements. The equivalent of an electret layer does not exist for solid-state MEMS elements, so they require an external source to bias the capacitive element.
The microphone element constitutes a purely capacitive load, so no current is drawn from the biasing reference. In an extended version of the amplifier system an on-die charge pump is included, thus addressing this problem by providing the necessary biasing for the microphone MEMS element.
Good design practices that reduce noise corruption in portable audio apps
Silicon MEMS microphones simplify audio design
Akustica launches HD microphone for mobile devices
Silicon MEMS microphone resources