Texas Instruments Audio and Video/Imaging Series
Click modulation is a coding technique that was developed in the mid 80's by Logan (1984) to retrieve information encoded by the zero crossings of certain bipolar signals. Somewhat akin to pulse-width modulation techniques, its main advantage is that the designer can specify a guard band that is free of distortion caused by intermodulation products. Therefore, the noise caused by the modulation technique can be pushed to a “don't care” band, mitigating the base band and guard band of deleterious distortion effects. Furthermore, the demodulation process can be easily performed by a simple low-pass filter. These advantageous features have been exploited by Streitenberger et al. , ,  to implement very low distortion switching (class-D) audio amplifiers. However, the main drawback of this technique is the extensive signal processing required to codify a given signal.
Contrary to PWM techniques, click modulation does not require a high switching frequency to avoid distortion, thus you can use a switching rate slightly higher than the Nyquist frequency. However, you must precisely determine the time instant where the switching occurs. A little uncertainty in the switching time will yield undesired distortion. To achieve low-noise levels, you should use a higher sampling rate in this PWM audio amplifier. Another option is the use of high-speed dedicated hardware  . Whatever hardware architecture is used to implement the click modulation algorithm in real-time, it requires a high-cost and high-power-consumption hardware that is not suitable for portable audio applications. A hardware prototype for a switching audio amplifier requires five digital signal processors with a maximum sampling rate of approximately 400 kHz ,  . Consequently, the power saving obtained by the use of high-performance, energy-efficient amplifiers is reduced by the requirements of high-speed signal processing hardware needed to implement the complex algorithm in real-time.
Nevertheless, many portable (and non-portable) audio applications (MP3 players, CD players, and so on) do not require real-time processing because they reproduce recorded music. Therefore, there is no need for a real-time implementation of the click modulation algorithm to take advantage of it.
In this article, we present an alternate off-line implementation of click modulation, where the coding algorithm is performed by software routines that yield a file of time-switching instants. This data is then fed to very simple hardware that generates a PWM-like output to directly drive a high-efficiency switching audio amplifier. In this way, a personal computer or processor can perform the coding procedure off-line, similar to the popular MP3 encoding procedure. After the encoded files are available, power-efficient hardware (such as the Texas Instrument's UCD9K family products) that does not perform complex signal processing can process the files and perform the usual tone and equalizer filters. The same hardware would drive a Class-D amplifier yielding a hi-fi, low-cost and high performance switching audio amplifier suitable for portable audio applications.
First, we compare PWM versus click modulation schemes to reveal the important advantages of the click modulation over PWM. Then, we present an off-line implementation of the click modulation algorithm. Simulation results clearly show the advantages of this modulation scheme over PWM. Lastly, we demonstrate a hardware implementation to reproduce the recorded audio in a digital audio amplifier. (For a brief introduction to click modulation, read Introduction to Click Modulation.)
The authors believe that this new off-line click modulated coding concept can play an important role in the portable, high-fidelity digital audio amplifiers.
Click modulation can be thought of as a modified PWM process  or a modified phase-modulation for single-sided band (SSB)  . For a given analog input signal f(t), the click modulation algorithm requires precise detection of the zero crossings of an intermediate signal s(t) to produce a switched output signal q(t) with the same low-frequency spectrum of f(t) and a separate carrier band. However, click modulation is better than PWM since the base and the carrier bands are completely separated and free of intermodulation distortion. The switching frequency should be somewhat larger than twice the maximum frequency of the audio signal, but is chosen as a tradeoff between cost and complexity of the output low-pass filter and the switching losses. Comparatively, to achieve similar distortion levels, PWM techniques require much higher switching frequencies in the range of 250 to 500 kHz for some commercial digital audio amplifiers.
The application of click modulation to audio power amplifiers was proposed by Streitenberger et al  , and the first practical implementation in hardware was later demonstrated by Streitenberger et al  . This modulation scheme is promising because of the isolation of the audio band from the distortion introduced by the modulation process, and the low switching frequency needed (compared to that of a typical PWM) to achieve low distortion. However, as reported in ,  up to five digital signal processors (DSP) are needed to implement a real-time click-modulated audio-power amplifier.
An off-line approach, as presented here, can use different solutions (finer sample rate, root finding via polynomial approach, and so on), that are performed by software, and therefore, the only drawback is the longer processing time. Nevertheless, each piece of music needs to be processed only once and its format stored for future use. The proposed click modulator is shown in Figure 1 .
The click modulation scheme shown in Figure 1 was simulated using MATLAB. The input signal f(t) is comprised of three sinusoidal tones whose frequencies are:
f1 = 2.5 kHz
f2 = 5 kHz
f3 = 10 kHz.
The carrier frequency is fP = 12 kHz. In these simulations, we did not use a special zero-crossing detector, rather we simulated the system using a high sampling frequency (fs = 10.24 MHz) to achieve a better precision in the zero-position detection of s(t). The sampling frequency fs determines the precision of the pulse positions, but it is independent of the switching frequency of the output signal q(t), which is equal to 2fP . In actual implementations, you can use smarter zero-crossing detection algorithms; thus, avoiding the need of these high sampling frequencies.
In this example, the signal f(t) to be encoded is bandlimited, allowing the use of a simple low-pass filter to recover the original signal. The corner frequency of the low-pass filter was chosen fc = 11 kHz. The filter gain at frequencies f > 13 kHz is very small, but not zero. Therefore, fU can be chosen as fU = 13 kHz, satisfying fP > (fs + fU )/2.
The waveform of f(t) is depicted in Figure 2a while the binary signal q(t) is shown in Figure 2b . The frequency spectrum of q(t) is shown in Figure 2c . You can clearly distinguish the three input signal tones, as well as the intermodulation tones due to the click modulation. Note that there are no spurious frequency tones over the frequency range between 0 and 12 kHz. The guardband extends from 10 to 12 kHz. The noise floor of -60 dB is caused by the aliasing effects at the output of the analytic exponential modulator, and to the resolution of the zero-crossing detections of s(t). In this case, the resolution is given approximately by ΩP /(πfs ) ≈ 1/427 ≈ -53 dB. We can further reduce this noise level using signal processing techniques such as oversampling, and/or using better zero-crossing detection algorithms  .
We require fast and accurate hardware (not high switching frequencies) to produce the exact pulse width. Commercial circuits, such as the Texas Instruments' UCD9K Fusion Digital Power Controllers, are capable of producing a digital PWM signal up to 150-ps resolution, reducing the noise floor down to -100 dB in this example. In a future article, the authors will analyze the effect of time resolution and the dead times upon the frequency spectrum of the synthesized waveform. An algorithm to linearize the Class-D amplifier (including dead times) was introduced in  . This algorithm could be run on the same digital power controller. Therefore, a combination of the off-line click modulation and the linearized amplifier would yield a power-efficient, high-fidelity digital audio amplifier.
Figure 3 shows a comparison between click modulation and natural PWM. To facilitate the visualization of the results, the signal f(t) is comprised of a single sinusoidal tone at 5 kHz. In both cases, the switching frequency is 2fP = 24 kHz, but the pulse widths are different. The frequency spectra of both signals are also shown in Figure 3 . Once again, the difference between the two modulation techniques is revealed. While the click modulation has no spurious content in the frequency band [0, fs ], the frequency spectrum of the natural PWM signal exhibits distortion tones in this band (at approximately 0.9, 3.8, and 9.0 kHz). For frequencies above fP = 12 kHz, the output q(t) of the click modulator shows a larger harmonic content than the output qPWM (t) of the natural PWM modulator. Therefore, click modulation achieves the separation between the signal band and the switching harmonics at the expense of a higher distortion outside the frequency band of interest.
Many portable audio applications (MP3 players, CD players, and so on) do not require real-time processing because they reproduce recorded music. Therefore, there is no need for a real-time implementation of the click modulation algorithm. The coding algorithm can be performed off line using software routines to yield a stored file of time-switching instants. The algorithm needs to be performed only once for each piece of music, and the data file can be stored for later use in the audio player (we called this device, the click amplifier). The general block diagram of a click amplifier (Figure 4: ) comprises:
- A digital input that reads a table from a data file containing the switching times of the signal across the load (signal q(t) in Figure 2 ) and stores it into the click amplifier's memory
- A digital PWM unit (DPWM) that converts each entry of the table to a signal that drives the gates of the H-Bridge switches
- An analog low-pass filter to suppress the high-frequency components.
The DPWM unit could be implemented using the TI's Digital Fusion family products (such as the UCD9501), followed by a MOSFET driver (UCD7201). The H-Bridge and the low-pass filter form a typical Class-D amplifier. However, you may consider designing a dedicated integrated circuit chip to lower the chip cost since this IC does not perform complex signal processing but only the usual tone and equalizer filter functions, besides the DPWM.
Different alternatives to generate the input data file are:
- This could become a new CD recording standard, thus, the music may already come in this format in the future
- The end user could run an application on his PC to convert from any recording standard (for example, CD, MP3, and so on) to the data file format needed for the click amplifier.
The size of the data file is not an issue, since the only information that needs to be stored is the starting and ending times for each pulse (which are generated at a rate of one every switching period). Nevertheless, as the price of digital memory is decreasing with each line-width reduction in the microelectronic fabrication industry, additional memory requirement does not necessarily increase the cost of the click amplifier.
A pre-processing click modulation scheme for digital audio amplifiers is presented. It is shown that this new digital format can improve the high-fidelity requirements needed for many portable applications, and, at the same time, reducing the power consumption. The new encoder moves the signal processing to an off-line personal computer or processor, reducing the processing power needed in the audio device. Therefore, power efficient, high-end audio quality can be achieved in these digital audio amplifiers.