Audio-power-amplifier systems have traditionally used linear amplifiers, which are well known for being inefficient. A linear Class AB amplifier is designed to act as a variable resistor network between the power supply and the load. The transistors operate in their linear region and voltage that is dropped across the transistors (in their role as variable resistors) is lost as heat, particularly in the output transistors. Figure 1 shows a Class AB amplifier configured in a bridge-tied-load, BTL, configuration.
Figure 1. Class AB amplifier
Class D amplifiers were developed as a way to increase the efficiency of audio-power-amplifier systems. The Class D amplifier works by varying the duty cycle of a pulse-width-modulated, PWM, signal. The amplifier increases duty cycle to increase output voltage and decreases duty cycle to decrease output voltage by comparing the input voltage to a triangle wave. The waveforms of a traditional Class D amplifier are shown in Figure 2.
The output transistors of a Class D amplifier switch from full OFF to full ON (saturated) and then back again, spending very little time in the linear region in between. Therefore, very little power is lost to heat. If the transistors have a low ON resistance little voltage is dropped across them, further reducing losses. The ideal Class D amplifier is 100 percent efficient, which assumes that both the ON resistance, RDS(ON), and the switching times of the output transistors are zero.
Fig. 2. Class D amplifier waveforms
ADVANTAGES OF CLASS D AMPLIFIERS OVER CLASS AB AMPLIFIERS
The advantage of Class D is increased efficiency. Increased efficiency means lower supply current, longer battery life and less heat dissipation.
Class D amplifiers have lower supply current:
Class D amplifiers can drive the same output power as a Class AB amplifier using less supply current. This is a major benefit for systems with a limited power supply. For example, the USB port supplies a 5-V, 500-mA power supply. This supply can be used to power a 2-W Class D amplifier, but a Class AB amplifier requires more supply current to drive 2 W. Thus, a USB speaker can get by without an external power supply while a Class AB amplifier requires an external power supply. The Class D solution not only reduces cost, but adds value through ease of use.
To show that the supply current of Class D is less than that of Class AB, the average supply current was measured playing music for both Class D and Class AB amplifiers. For this test we set the Class AB and the Class D amplifiers to the same gains and set the volume setting of a CD player the same for each amplifier. As the song played, a Data Acquisition Card was used with a computer and was set to sample the current 44,000 times per second and calculate the average supply current throughout the song. The computer also sampled the output voltage and found the peak and RMS voltages of the song. These voltages enabled the crest factor (crest factor = peak output voltage / average output voltage) of the song to be calculated. It was 15dB. The test was repeated several times increasing the output one step in volume each time. The test was run over normal listening levels with the amplifiers driving 8-W speakers from a 5-V supply.
Figure 3. Average supply current vs. peak output voltage
Class D amplifiers offer longer battery life to portable devices:
Increased efficiency allows portable devices that use Class D amplifiers to last longer. To show this, battery life was measured for two equal systems: one using a stereo Class AB amplifier and one using a stereo Class D amplifier. Using a standard Rayovac 9-V battery with a 9-V-to-5-V dc-to-dc converter in each system and routing the same audio signal into both systems, we let it run and monitored the battery voltage. The dc-to-dc converter module automatically shuts down when the battery voltage drops below about 5.2 V under voltage lockout (UVLO). So, UVLO was used as an indicator that the battery was dead. The spikes you see on the battery voltage waveforms happen when the UVLO event occurs and the voltage drifts up until the load is reapplied. The system is reset, allowing three UVLOs to ensure that the battery is dead. Figure 4 shows that the Class D amplifier lasted 2.5 times longer than the Class AB amplifier.
Fig. 4. Battery life test of Class D and Class AB amplifiers playing music
Class D amplifiers dissipate less heat:
Getting heat out of a system is becoming more important as devices get smaller and processors run faster. Class D amplifiers dissipate less heat than Class AB amplifiers, because Class D amplifiers are more efficient. Class AB amplifier output stage operates as a resistor divider, causing power to be dissipated in the output transistors, while a Class D amplifier switches between fully "off" and fully "on" so little power is dissipated in the output stage. The maximum ambient temperature depends on the heat-sinking ability of the package and power dissipation in the amplifier.
To calculate maximum ambient temperatures, first consider power dissipation. The Class AB internal dissipation was calculated assuming an ideal Class AB amplifier, and the Class D power dissipation was calculated given measured efficiency versus output power. Given ((CK BETWEEN Q AND JA HERE AND BELOW)) Q_JA of 58.5 C/W, the maximum allowable junction temperature and the total internal dissipation, the maximum ambient temperature can be calculated with the following equation. The maximum recommended junction temperature for most audio amplifiers is 150C. Crest factor was calculated as a ratio of average power to maximum peak power.
Figure 5. Maximum ambient temperature versus crest factor for Class D and Class AB
Figure 5 shows that Class D amplifiers stay much cooler than Class AB for the same output power. In cases of high output power, the Class AB amplifier would require a heat sink to decrease Q_JA.
FILTERLESS CLASS D MODULATION SCHEME
Until now, the only thing keeping Class D audio amplifiers out of many applications was the large output filter that drives up cost and size. The filterless modulation scheme allows the designer to do without the filter while keeping the great efficiency advantages that Class D has over Class AB.
Typical Class D modulation scheme:
The typical Class D modulation scheme has a differential output where each output is 180 degrees out of phase and changes from ground to the supply voltage, VDD. Filtered, 50 percent duty cycle yields 0 V across the load. The differential prefiltered output varies between positive and negative VDD; therefore, there is always VDD across the load if not using a filter. The typical Class D modulation scheme with voltage and current waveforms is shown in Figure 6. Note that even at an average of 0 V across the load (50 percent duty cycle), the current to the load is high causing high loss thus causing a high supply current.
Figure 6. Typical Class D's output voltage and current waveforms into an inductive load with no input
Filterless modulation scheme:
The TPA2000D2 uses a modulation scheme that still has each output switching from 0 V to the supply voltage. However, OUT+ and OUT- are now in phase with each other with no input. As shown in Figure 7, the duty cycle of OUT+ is greater than 50 percent and OUT- is less than 50 percent for positive voltages. The duty cycle of OUT+ is less than 50 percent and OUT- is greater than 50 percent for positive voltages. The voltage across the load sits at 0 V instead of VDD throughout most of the switching period. This greatly reduces the switching current, which reduces any I2R losses in the load.
Figure 7: Filterless Class D's output voltage and current waveforms into an inductive load
ELIMINATING THE OUTPUT FILTER
This section will focus on why the user can eliminate the output filter with the filterless modulation scheme. The advantages and disadvantages of not using a filter will also be discussed. Any measurements of the filterless modulation scheme were made with the TPA2000D2.
Class D audio without a filter:
Audio specialists have said for years, "Do not apply a square wave to speakers." If the amplitude of the waveform is high enough and the frequency of the square wave is within the bandwidth of the speaker, the square wave could cause the voice coil to jump out of the air gap and/or scar the voice coil. A 250-kHz switching frequency, however, is not as significant because the speaker cone movement is proportional to 1/f2 for frequencies beyond the audio band 1. Therefore, the amount of cone movement at the switching frequency is very small.
Class D amplifiers put out a pulse-width-modulated (PWM) square wave, which is the sum of the switching waveform and the amplified input audio signal. The human ear acts as a bandpass filter so that only the frequencies between approximately 20 Hz and 20 kHz are passed. The switching frequency components are much greater than 20 kHz, so the only signal heard is the amplified input audio signal.
Efficiency with and without a filter:
The main reason that the traditional Class D amplifier needs an output filter is that the switching waveform results in maximum current flow. This causes more loss in the load, which causes lower efficiency. The ripple current is large for the traditional modulation scheme because the ripple current is proportional to voltage multiplied by the time at that voltage. The differential voltage swing is 2*VDD and the time at each voltage is half the period for the traditional modulation scheme. An ideal LC filter is needed to store the ripple current from each half-cycle for the next half-cycle, while any resistance causes power dissipation. The speaker is both resistive and reactive, whereas an LC filter is almost purely reactive. Therefore, the traditional modulation scheme is efficient with an LC filter, but not efficient without an LC filter.
The filterless modulation scheme has very little loss in the load without a filter because the pulses are very short and the change in voltage is VDD instead of 2*VDD. As the output power increases, the pulses widen making the ripple current larger. Ripple current could be filtered with an LC filter for increased efficiency, but for most applications, the filter is not needed.
An LC filter with a cutoff frequency less than the Class D switching frequency allows the switching current to flow through the filter instead of the load. The filter is less lossy than the speaker, which causes less power dissipated at high output power and increases efficiency in most cases. The LC filter referenced to ground uses L1 = L2 = 22 uH and C1 = C2 = 1 uF from Figure 8. This filter does not take advantage of the small differential pulses of the filterless Class D modulation scheme because it filters the output of OUT+ and OUT- independently. Therefore, the quiescent current is higher with the filter, because ripple currents increase flowing through the inductor to ground (like the traditional Class D modulation scheme).
If the designer is worried about EMI but does not want a large, expensive filter, ferrite beads could be used in place of the inductors. In this ferrite filter, L1 and L2 are chip ferrite beads and C2 and C3 are 1 nF capacitors. If capacitors C1 and C2 are too large, the cutoff frequency is reduced, but the efficiency is also reduced. This filters the high-frequency components that could cause EMI problems in the circuit and has a cutoff frequency between 5 and 10 MHz.
Figure 8. Class D output filter
Figures 9 and 10 show the measured efficiency of the 2-W filterless Class D amplifier, TPA2000D2, with various filter options and compare the efficiency to the efficiency of a Class AB amplifier 2. The supply current was measured with a Tek AM501 DMM. The output voltage was filtered and fed into an Audio Precision II and the output current was measured with a current probe. GC International inductors, SCD0703T-220M-S, were used in the LC filter. The inductors had an average dc resistance, DCR, of 110 milliohm. The high DCR was used to show that good efficiency could be obtained even with small inductors. Fair-Rite's 2512067007Y3 ferrite chip inductor was used with 1nF capacitors as the ferrite filter. An 8-ohm load was used in Figure 9. The 8-ohm speaker was used without a filter and with the ferrite filter, and an 8-ohm resistor was used for the LC filter. A 3.3-ohm load was used in Figure 10. A speaker used in notebook computers that were labeled 3.3 ohms was measured at 2.7 ohms at dc. The notebook speaker was used without the filter and with the ferrite filter. A 3.3-ohm resistor was used with the filter. The Class AB efficiency in Figures 9 and 10 is the calculated best-case Class AB efficiency with no bias current.
Figure 9 shows that the ferrite filter was the most efficient, and its efficiency increases faster than the LC filter with increased output power. The ferrite filter's high efficiency at max power is due to its low DCR (DCR < 50="" mw_).="" the="" tpa2000d2="" without="" the="" filter="" has="" higher="" efficiency="" at="" max="" power="" than="" with="" the="" lc="" filter.="" the="" efficiency="" of="" tpa2000d2="" without="" a="" filter="" is="" a="" balance="" of="" no="" dcr="" loss="" and="" increased="" switching="" loss="" in="" the="" load.="" all="" class="" d="" curves="" are="" much="" more="" efficient="" than="" the="" class="" ab="" amplifier.="">
Figure 10 results agree with Figure 9's except that the filterless application is less efficient than the LC filter application at the high output powers in Figure 10. The TPA2000D2 with LC filter was more efficient than without the filter because switching loss in the load for the filterless case was greater than losses in the DCR of the inductor. The maximum efficiency of an amplifier driving an 8-ohm load is greater than an amplifier driving a 3.3-ohm load because the resistive losses are much more significant with a smaller load.
Fig. 9: Efficiency of the TPA2000D2 with various filters into an 8-ohm load compared to the efficiency of a Class AB amplifier
Figure 10: Efficiency of the TPA2000D2 with various filters into a 3.3-ohm load compared to the efficiency of a Class AB amplifier
Power requirements for speaker:
As discussed previously, the 250-kHz square wave will not damage the speaker by causing the voice coil to jump out of the air gap. However, the speaker could be damaged if the voice coil is not designed to handle the additional power. To size the speaker for added power, the ripple current dissipated in the load has to be calculated by subtracting the theoretical supplied power, PSUP THEORETICAL, from the actual supply power, PSUP, at maximum output power, POUT. The switching power dissipated in the speaker is the inverse of the measured efficiency, h_MEASURED, minus the theoretical efficiency, h_ THEORETICAL.
The theoretical efficiency of the TPA2000D2 with an 8-W_ load is 88 percent (RDS(ON) = 0.5 W_)_. Given a maximum output power of 1.3 W, and a maximum measured efficiency of 78 percent, we calculate using equation 4 that there is an additional 190 mW dissipated in the speaker. The added power dissipated in the speaker is not a problem as long as it is taken into account when choosing the speaker.
Radiated electromagnetic interference (EMI) is radiation caused by the transfer of electromagnetic energy through a nonmetallic medium, such as air. EMI is caused either by an instantaneous change in current resulting in a magnetic (H) field or by a voltage resulting in an electric (E) field. Devices that are required to meet a certain level of EMI performance, like FCC and CE approval, usually test far field E field and not H field. The H field primarily causes problems in the near field, which could cause surrounding circuitry inside the device to fail due to inductive coupling. The inductor in the filter, or the inductance in the speaker if not using a filter, keeps the change in current low, which decreases the magnetic field in the speaker wires. However, the inductor also acts as an EMI radiator. The electric field, which is a common-mode concern, could be quite large because of the switching voltage if the speaker wires are long enough.
The LC filter used in the efficiency graphs has a common-mode filter with a cutoff frequency less than the switching frequency of the amplifier, so the E and H fields are attenuated. If EMI is a problem without a filter, the designer can use a ferrite bead filter to reduce the EMI at a much lower cost than using an LC filter. However, the TPA2000D2 evaluation module passed FCC and CE without any output filter with speaker wire lengths of eight inches and less.
The E and H field output traces near the TPA2000D2 were measured with no filter, a capacitor to ground, a ferrite bead filter(mentioned previously) and an LC filter (mentioned previously). The tests used an FFT function oscilloscope and EMCO E and H field probes (EMCO 7405-904 and EMCO 7405-901). The circuit was configured with both channels of the TPA2000D2 EVM active with no input to see the EMI generated by the switching waveform. An 18-inch speaker wire connected the EVM to the speaker. Each speaker was spread apart away from the EVM board to ensure that the EMI from the traces did not add to the EMI measured from the speaker wire and vice versa. The measurements were done in the near field, one inch above the output traces. The FFT function was used to see the effect of filtering.
Figure 11 shows the E field measured one inch above the output traces for the TPA2000D2 with no filter, a 470-pF capacitor to ground, a ferrite bead filter and LC filter. Each spike on each of the curves represents a harmonic of the 250-kHz switching frequency. As shown in Figure 11, the TPA2000D2 without the filter radiates the most. The E field radiation was slightly reduced for higher frequencies with a 470-pF capacitor to ground. The ferrite bead filter attenuated the E field for frequencies greater than 10 MHz and there was even less radiation than the LC filter for frequencies greater than 25 MHz. E field radiation less than 30 MHz is usually acceptable because FCC and CE test radiated emissions only for frequencies greater than 30 MHz. The E field of the LC filter was very small for all frequencies.
Figure 11. E field one inch above output traces: no filter, 470-pF output to ground, ferrite filter and LC filter
Figure 12 shows the H field measured one inch above the output traces for the TPA2000D2 with no filter, a 470-pF capacitor to ground, a ferrite bead filter and an LC filter. The TPA2000D2 without filter radiates the most. The H field radiation was slightly reduced for higher frequencies with a 470-pF capacitor added from the outputs to ground. The ferrite bead filter attenuated the H field for frequencies greater than 10 MHz and it was the best at reducing the H field. Although the LC filter had a small E field, the H field of the LC filter was very large because the inductors radiate.
Fig. 12. H field one inch above output traces: no filter,470-pF output to ground, ferrite filter and LC filter
For most systems, the ferrite bead filter is the best solution for EMI. This is because low H field is important so signals from other circuitry are not corrupted and the E field radiation is usually a concern only for frequencies greater than 30 MHz.
SYSTEM LEVEL DESIGN ISSUES FOR CLASS D AMPLIFIERS
Although Class D amplifier system-level design is much easier than one might think, there are still some concerns that a designer must think about to optimize performance. These are discussed below.
Choosing a Class D output filter:
Cost is a primary concern with most products today. Therefore, if you don't need a filter, why use one? An LC filter costs approximately $1 for a 2-W stereo Class D amplifier. With the filterless modulation scheme the filter can be eliminated as long as EMI is not a problem. The TPA2000D2 EVM passed FCC and CE radiated-emissions tests for speaker wires less than eight inches. So, if you can get by with speaker wires shorter than eight inches, don't use an output filter.
But even if the speaker wires are longer you still might not need a full LC filter with the filterless modulation scheme. A ferrite bead filter mentioned above will work for most cases and is very inexpensive. If using a ferrite bead filter, make sure that the DCR of the ferrite bead is small so output power and efficiency are not limited.
If all else fails, use an LC filter. If the final product is very EMI-sensitive to low frequencies, an LC filter is required. It is required for traditional Class D modulation schemes--only the filterless modulation scheme can be used without an output filter.
Power supply decoupling:
Decoupling capacitors serve to smooth the supply voltage and assist the amplifier by providing current when needed. They may shunt relatively large ripple currents and must have a low equivalent series resistance (ESR) to reduce power and heat dissipation in the device. The ESR combines all losses, both series and parallel, in a capacitor at a given frequency in order to reduce the equivalent circuit to a simple RC series connection, valid only for low frequencies (less than 1 MHz).
Other considerations are the voltage rating, capacitance, physical size and the specific type of capacitor. The voltage rating should exceed the maximum supply voltage expected in order to handle voltage surges and spikes without being damaged. The capacitance is then important, as it specifies the amount of energy that can be stored in the capacitor.
Large capacitors typically do not act as capacitors at high frequencies, so it is very important to use both ceramic chip capacitors and large bulk capacitors. A small chip capacitor ( usually 0.1 uF to 1 uF) should be used for each supply pin of the amplifier and placed as close to the pin as possible to handle the high switching frequency of the Class D amplifier. A large bulk capacitor should be used for the amplifier to handle low frequency and high current of the audio signal.
Careful board layout ensures not only proper operation, but also good performance. The most important thing is power dissipation. If using a PowerPAD device, where the ground plane is used to get the heat out of the device, make sure to make the ground plane as large as possible. The larger ground plane lowers Q_JA, which lowers the temperature and increases the performance of the device.
Keep PVDD and VDD separated back to supply if possible. (PVDD is the supply voltage pin used for the power MOSFETs; VDD is the pin used for supplying power for the sensitive analog circuitry.) Switching current and high output current causes the voltage on PVDD to vary. If PVDD and VDD are kept separate back to the supply, the changes in voltage will be filtered and the voltage on VDD will remain more constant.
Keep decoupling capacitors close to power pins. The trace connecting the supply pin to the decoupling capacitor is a resistor and inductor that increase as the length of the trace increases.
Class D amplifiers are two to four times more efficient than Class AB amplifiers, which leads to longer battery life, less heat and lower supply current. Traditional Class D amplifiers have a large and expensive output filter; however, filterless Class D amplifiers eliminate the output filter without sacrificing performance. If EMI is a problem a ferrite bead filter can be used with the filterless modulation scheme, which is much smaller and cheaper than an LC filter. Power supply decoupling is very important for Class D amplifiers because the decoupling capacitors supply extra switching current and smooth the supply voltage. Board layout is key for optimizing performance of a Class D amplifier--little things like separating analog VDD and power VDD can greatly improve performance.
1 Colloms, Martin, High Performance Loudspeakers, Pentech Press Ltd., London, 1985, pp.18--26
2 TPA2000D2 2-W Filterless Stereo Class D Audio Power Amplifier Datasheet, Texas Instruments Inc., March 2000, publication number SLOS291B.
3 Score, Michael, Reducing and Eliminating the Class D Output Filter, Texas Instruments Inc., August 1999, publication number SLOU023.
4 TPA2000D2 Filterless Class D Audio Power Amplifier Evaluation Module Users Guide, Texas Instruments Inc., March 2000, publication number SLOU077.