Buzzwords such as 'capacitorless' or 'cap-free' are often heard today in connection with headphone amplifiers. Several such solutions exist on the market today, based on radically different technologies. Their relative benefits and drawbacks are not always obvious – ironically, some of the most attractive solutions actually need a higher number of capacitors than the time-honoured traditional circuit, but score instead on metrics like power consumption, pop suppression and startup time. This article provides an in-depth look at the issues and the available options.
What's the problem with capacitors?
A traditional headphone drive circuit is shown in Figure 1. The left and right output amplifiers operate on a single supply voltage, VDD, and the DC voltage at its outputs is mid-rail, i.e. VDD/2. Two capacitors are inserted after the amplifier to remove this DC voltage.
Figure 1: Traditional headphone drive circuit
Electrolytic or tantalum capacitors are typically used, one very common capacitance value being 220uF. Their capacitance and the headphone impedance together determine the circuit's response to bass signals; tones below the cut-off frequency (fc) are attenuated. With 220uF capacitors and 16 Ohm headphones, the cut-off is at 45Hz, falling to 22.5Hz if 32 Ohm headphones are used instead. Capacitance values below 220uF are undesirable as they push the cut-off frequency upwards, leading to a loss of bass content which even today's advanced signal-processing techniques can only partially redress.
While capacitor technology continues to evolve and improve, it lags behind the overall pace of miniaturisation and cost reduction in consumer electronics, which is driven by Moore's Law. The result is that today, a pair of 220uF capacitors simply takes up too much space on a personal media player or mobile phone circuit board. Although trade-offs can be made between the capacitors' physical size, height and cost, the traditional solution is ultimately unsatisfactory for most applications now. This is the primary issue with the circuit of Figure 1.
Another, less obvious, problem occurs when the circuit starts up. Before start-up, all circuit nodes are at 0V and both capacitors are uncharged. But during normal operation, the left side of each capacitor is at VDD/2 (in DC terms) while the right side stays at 0V. To get to this state, the capacitors must be charged by driving a DC current through them. If the amplifier outputs jump instantaneously from 0V to VDD/2 during startup, this charging current takes the form of a large, short current spike. Since any current through the capacitors also passes through the headphones, this produces an extremely loud 'pop' noise, which is unacceptable in the market today. Bringing the amplifier outputs up more slowly can reduce the amplitude and slew rate of the charging current to the point of making it virtually inaudible, but only at the cost of drastically increasing startup time. This is a significant drawback because audio playback is often used as part of the user interface, e.g. to confirm that a button has been pressed or an option selected. Long delay times between such user input events and the expected confirmation tone will make the system appear sluggish.
Both pop noise and user interface responsiveness are critical to the end-user experience, leaving system designers caught in a dilemma. Little surprise, then, that many choose to avoid the startup scenario altogether by simply leaving the headphone amplifiers powered up ” even when they are not needed. This practice increases standby power consumption and goes against the philosophy of rigorous and fine?”grained power management, which has become the norm in battery?”powered systems. Audio component vendors have responded by providing low-power standby modes that keep the amplifier outputs biased at VDD/2, while consuming less power than during playback. However, this is only a partial solution as it relies on the VDD supply voltage remaining available – i.e. the voltage regulator that generates VDD cannot be shut down, reducing standby battery life.
To sum up, the traditional headphone drive circuit forces system designers into compromises that are becoming increasingly unacceptable – firstly, a three?”way trade?”off between the capacitors' physical size, their cost, and the system's bass response; and secondly, a painful choice between pop noise, a long startup time, high standby power consumption and adding extra components.
An alternative circuit without capacitors is shown in Figure 2. Here, a third amplifier channel has been added and connected to the headphones' ground terminal (the 'sleeve' on a typical TRS connector). It acts as a 'virtual ground', producing a DC voltage of VDD/2, with no AC component. The left and right channels are unchanged from the traditional circuit of Figure 1. Since the DC voltage between either left or right and the virtual ground is zero, there is no need for DC blocking capacitors.
Figure 2: Capless headphone drive circuit using a virtual ground
The benefits of this solution are threefold. Firstly, it can be built smaller, lower, and cheaper than the traditional circuit; secondly, its bass response is flat, allowing for accurate reproduction of bass content; and finally, startup is less problematic because there is no need to charge DC blocking capacitors. Component vendors have offered the 'virtual ground solution for years using marketing names such as 'pseudo-differential', 'output capacitor-less' and 'phantom ground'. It has been used by many OEMs, including well-known brand names.
However, the solution is not without problems. One of its drawbacks is the power consumed in the additional 'virtual ground' amplifier. Assuming Class B amplifiers with a small output amplitude and a resistive load, this is equal to the left and right channel amplifiers combined – i.e. the circuit's total power consumption is twice that of the traditional circuit under similar conditions. Even with a full-scale sine wave signal, the 'virtual ground' solution still consumes 64% (2/pi) more power than the traditional circuit. The result is a significant reduction in playback battery life, at any volume level.
Another problem can arise when headphone sockets are used as a line-out. In portable systems without separate line-out sockets, end users often connect the headphone output to the line input of a home hi-fi system or docking station, using commercially available adaptor cables. Depending on the grounding at both ends (the portable system may be grounded through a charger), the virtual ground could then be directly connected to a true ground. The resulting short circuit would prevent the audio signal from being transmitted correctly. With headphone amplifiers engineered to withstand short circuits of any duration, the risk of permanent damage should be remote ” but even so, it's clearly not ideal from a reliability point of view. Overall, 'virtual ground' is a useful alternative to the traditional circuit in many applications, but has its own drawbacks that prevent it from becoming the industry standard solution.
Solving the problems of the traditional headphone drive circuit without introducing new issues requires 'ground-referenced' amplifiers whose output voltage is centred around 0V. Such amplifiers need a symmetric supply consisting of one positive and one negative voltage of equal magnitude. Since negative supply rails are rare in consumer electronics systems, a number of component vendors have integrated charge pumps in their audio ICs, as shown in Figure 3. This solution is being promoted under several different trademarks.
Figure 3: Ground-referenced headphone drive circuit
Ironically, this solution actually needs more capacitors than the traditional circuit: one at the input to the charge pump, one at its output, and a 'flyback' capacitor. (The capacitor at the input sometimes hidden away in the small print of IC datasheets – it is needed to compensate for the non-ideal transient response of real-world power supplies, not directly due to the charge pump itself). Clearly, 'capless' is not the most appropriate description for this circuit. Its main benefit lies in the overall size and cost of external components. Being in the single digits of microfarads, these three capacitors compare very favourably with the two 220F capacitors of the traditional circuit. Unlike virtual ground solutions, the circuit provides a true ground output and can be used as a line output without restrictions. It can also run from lower supply voltages, as the inverting charge pump doubles the amplifier's voltage swing.
The key remaining issue with ground-referenced solutions is power consumption. The efficiency of charge pumps is limited at low volumes by switching losses, and at high volumes by interconnect resistance and the physical size of switching elements on the chip (enlarging them increases cost). Moreover, some amplifier designs cannot tolerate the supply ripple generated by the charge pump, leading several vendors to add LDO voltage regulators to remove this ripple. The regulators' dropout voltage causes further losses. Overall, the power efficiency of most ground-referenced solutions is only about half that of the traditional circuit, shortening playback battery life.
Overcoming the power efficiency limitations of ground-referenced headphone drivers is becoming a hot topic in the low-power audio community. Class-G amplifier architectures, where the amplifier's supply voltage is adjusted depending on the volume of the audio signal, are a particularly promising avenue. However, standard inverting charge pumps with their fixed output voltage cannot support Class-G. Wolfson Microelectronics' first ground-referenced Class-G device, the WM8900, solved this problem with a novel charge pump design with two inputs. These are connected to two different supply voltages that are available in the majority of battery-powered equipment today, enabling the charge pump to generate two different output voltages.
A more recent Wolfson development, dubbed 'Class-W', is incorporated in the WM8903 audio codec. Here, the charge pump has only one input – typically connected to an existing 1.8V supply rail – but two outputs, VPOS and VNEG, which provide a symmetric supply to the amplifier. The magnitude of VPOS and VNEG changes to suit the signal volume, improving power efficiency in much the same way as other Class-G implementations. However, Class-W goes beyond this by also adjusting the charge pump's switching frequency depending on signal volume. This reduces switching losses and further increasing battery life. Compared to standard inverting charge pumps, this design requires just one additional pin and a small additional capacitor, as shown in Figure 4.
Figure 4: 'Class-W' headphone drive circuit
The WM8903 headphone amplifiers have a high PSRR (power supply rejection ratio), and can therefore run directly from the charge pump, without any need for on-chip LDO regulators. Other analogue circuitry within the device is also engineered for high PSRR, eliminating the need for external LDOs in many cases. A variety of other, unrelated power-saving techniques are used in the digital core, digital-to-analogue converters and other parts of the device.
While older solutions will continue to be used for some time, the new industry standard for headphone drivers in handheld applications is ground-referenced. Additional must-have requirements include ultra-low power consumption – in the single digits of milliwatts for playback under real-world conditions – small solution size including external components, no compromise in audio quality, and reasonable cost. As component vendors strive to meet these tough demands, new solutions and incremental improvements will continue to appear. In all likelihood, some will be trade-offs where some design goals are achieved at the expense of others. Solutions which succeed in achieving all key criteria at the same time, are likely to be adopted on a massive scale.