Technical & Editorial
- Written by Dr. David A. Rich
- Published on 29 January 2014
- AVR - Audio Video Receiver - Build Quality: Part II - Design of High-Performance AVRs and Pre/Pros
- Page 2: Electronic Volume Controls that Enhance Performance
- Page 3: Operational Amplifier Selection
- Page 4: Differential and Current Mode DACs
- Page 5: Multiple DACs Combined to Produce a Single Channel Output (Mono Mode)
- Page 6: Quasi Current-Mode Interface for ESS DACs with a Single Operational Amplifier
- All Pages
Design of High-Performance AVRs and Pre/Pros: A More Technical Perspective with Material Relevant to Stereo Preamplifiers and Outboard DACs
Following on Part I of this series, "AVR - Audio Video Receiver - Build Quality: Part I", here, in Part II, I dig a bit deeper into the performance of a single-chip Large Scale Integrated (LSI) analog AVR chip and introduce better solutions employing Small Scale Integrated (SSI) analog chips. Herein, I refer only to Pre/Pros, but everything discussed applies equally to AVRs. Pre/Pros tend to use analog SSI chips, while most AVRs use the LSI solution (see the chart in Part I of this article for the make and model numbers of the AVRs and Pre/Pros using the LSI chip). I carryover the term, AVR LSI, from the first part of this article and refer to AVR when I comment on products with power amplifiers.
The sections on electronic volume controls, op-amps, and CMOS-based signal switching are also relevant to the stereo preamplifier design. The section on circuitry interfacing to a DAC, digital reconstruction filtering, and jitter attenuation are integral to stereo devices with digital inputs.
Electronic Volume Controls that Enhance Performance
While AVR and pre-pro marketing focuses on the highest-resolution component – usually the DAC chip - an audio signal chain is ultimately limited by its lowest-resolution circuitry. Often, that is the volume control. The difference between the THD of the best IC-based volume control ICs (Integrated Circuits) and typical AVR LSIs (Large Scale Integrated Circuits) is 2.5 bits using measurements sourced from data sheets with the control set to unity gain.
In a traditional volume control, a fixed gain stage [typically 20-26 dB of gain] follows a voltage attenuator (potentiometer). The fixed-gain stage introduces significant noise and distortion. For most of the range of the volume control, the attenuation exceeds the gain. Under these conditions, the gain following the volume control is unnecessary.
In some older SSI (Small Scale Intgration)-based multi-channel products and stereo preamps using an electronic volume control, an obsolete IC was used. This old IC contained only one resistor ladder and was a direct replacement for the potentiometer. This part has been discontinued, favoring a more modern design discussed below.
In a modern electronic volume control, two resistor ladders are integrated on the chip. One is placed in the traditional position before the amplifier. The other resistor string is used to adjust the gain of the amplifier. With this topology the amplifier is wired as a unity gain buffer (0 dB), and has much lower noise and distortion than an amplifier required to provide a gain of 20-26 dB).
Only when the attenuator before the volume control moves to the unity gain position does the second resistor string become active to increase the gain of the amplifier above unity. The added gain, with its associated noise and distortion, only occurs when it is needed. The complete circuit is called a "Gain Optimized Volume Control". Since this circuit requires two resistor strings, whose dual operation must be coordinated, it is not realizable using a passive volume control.
An SSI option for high-performance volume controls consisting only of resistor ladders is readily available now. The latest part introduction is from New Japan Radio. The part (the MUSES 72320) is found in the $1000 stereo Emotiva preamplifier reviewed here in Secrets.
The JRC MUSES 72320, or the similar, recently introduced, Micro Analog Systems MAS6116, has four resistor strings to allow for one chip to be used in a stereo system. These chips are not used in current Pre/Pros whose internal design I am familiar with. Instead, high-performance, dynamic range optimized, electronic volume controls in Pre/Pros with SSI designs are from Cirrus and TI. In addition to the resistor ladders, these parts include an operational amplifier (one per channel) internal to the chip. The chips are available with two to eight sections. These parts started to appear in the early part of the last decade.
The eight-section Cirrus CS3318 consumes the same amount of current as the LSI AVR chips, despite having fewer than half the op-amps. Since the Cirrus 3318 only functions as a volume control, each op-amp consumes more silicon area. The Cirrus operates at a higher power supply voltage of ± 9V. This all translates into better performance. Using typical numbers, at unity gain, the Cirrus data sheet reports distortion as 0.00025% THD+N, while the data sheets for the LSI AVR chip report 0.001% for the same output signal swing. This is a 12 dB improvement at 1 kHz. In bits equivalent, we shift from 16.5 bits to 18.5 bits. With eight channels of volume controls, the Cirrus is more accurately called a "Medium Scale Integrated" Circuit (MSI).
The TI PGA4311UA is a four-channel component that has lower distortion than the Cirrus between 600 mVRMS and 2 VRMS. Two of the TI chips provide the volume control function for the eight channels. Since twice the TI chips are required, the pair is more expensive than the eight-channel Cirrus part. Worst-case distortion of the TI part with an output voltage of 2 VRMS is 0.004%. A single chip AVR LSI has a specified worst-case distortion of 0.02% at 2 VRMS at its output. This is five times the distortion of the TI PGA4311UA.
Secrets' Audio Precision measurements of multi-channel products using two versions of the LSI AVR chip (Renesas and Rohm) and the Cirrus CS3318 confirms the Cirrus chip has lower distortion at 1 VRMS. The distortion vs. frequency measurement stays flat with the Cirrus chip, but the Renesas starts to rise about 2 kHz, moving to four times the mid-band value at 20 kHz. Complete THD+N vs. Frequency was not available for the Rohm chip. The SNR measurements of the AVR with the CS3318 show at least a 1/2 equivalent bit improvement in comparison to the Renesas and Rohm chips.
Electronic volume control performance can be improved with two volume controls sections in balanced mode, but does so at twice the cost.
Balanced mode minimizes distortion and expands dynamic range of the volume control. The op-amp at the bottom left inverts the incoming signal and the Single Ended to Balanced Converter restores the signal to unbalanced mode. The balanced-to-single-ended converter consists of resistors and one op-amp. The Emotiva XSP-1 stereo preamp, mentioned above, uses a balanced volume control topology.
Secrets' measurements of the XSP-1 preamp shows improved distortion performance over the single ended Cirrus CS3318. Distortion of the Emotiva at unity gain was -106 dB from 20 Hz to 20 kHz at 2 VRMS with almost not change at 5 VRMS.
In theory a Pre/Pro with balanced outputs could have a completely balanced signal chain from the DAC output through the analog volume controls to the balanced output connections at the back of the unit. However, practical issues, such as the balanced signals becoming mismatched as they travel through the signal path, present a roadblock. In addition, common mode signals that accumulate as the balanced signal moves through the unit are not removed.
The engineer has many choices for a fixed build of material when designing an electronic volume control. For example, the cost of using two sections of a Cirrus CS3318 in balanced mode might be similar to using the MUSES 72320 single ended with high quality op-amps.
None of my service manuals for multi-channel products pointed to the balanced technique. The technique was adopted in the B&K Reference 70 which, unfortunately, never made it to production. Designed by the extraordinarily talented Ed Mutka, the Reference 70 was truly innovative.
The most elegant configuration for an electronic volume control uses current-mode switching with bipolar devices. An example is the Texas Instruments DAC8812. The voltage at the current switch remains constant at ground potential, which introduces far less distortion than voltage-mode switching where a tap is selected on the resistor ladder.
The resistors on the Texas Instruments DAC8812 have an added layer of deposited thin film metal to reduce distortion, but this adds significantly to the price of the silicon. The resistors in the other volume controls are made from material already deposited to make an IC with MOSFET transistors. Typically, these are polysilicon wires that form the gate of the MOSFET. Distortion in polysilicon resistors occurs when the resistance changes slightly as the voltage across the resistor changes. This is called the voltage coefficient of the resistor, which is a specification sometimes found in data sheets for discrete resistors.
The current at the output of the selected switches is converted back to a voltage by an op-amp circuit functioning as a current-to-voltage converter. This method is used in high-end two-channel products by Accuphase and Simaudio Moon, although they may use other options than the TI chip mentioned above. It is unlikely an affordable Pre/Pro could include this chip. See the example in this link to the Accuphase C-2120 Stereo Control Center.
Operational Amplifier Selection
As mentioned in Part I, some SSI op-amps (1 – 4 in a package) must be used around the single chip LSI AVR chip for anti-alias filtering (ADC) and reconstruction filtering (DAC). Unless otherwise noted in the chart, these are minimal cost op-amps such as the LM833 and RC4558. They were among the first practical audio op-amps, dating to the 1970s. Many other op-amp part numbers are found in service manuals. These are slight improvements on the original op-amps, where the improvements arose from advances in semiconductor processing technology. The incremental part cost is several pennies.
For 50% more cost per op-amp, the NE5532 can be used. This is a much more complex chip, allowing it to have dramatically lower noise and distortion. Confusingly, the single op-amp version of this chip is given the part number NE5534
Some DAC vendors supply a Printed Circuit (PC) evaluation board for designers to verify the performance of the DACs. Achieving optimal performance is very dependent on associated passive and active parts as well as PC board layout. If the DAC producer has developed an evaluation board for a given part, a PDF on its design will be on the website of the IC Vender. These application documents often provide measurements of the complete board that are more comprehensive than the information provided in the DAC's datasheet for the DAC alone. The IC vendor's evaluation board documentation provides details on how to get the best performance from a specific DAC and is an excellent place to go for more technical detail than I can provide here. Not all DACs from an IC vendor have evaluation boards.
An evaluation board for the vendor's top of the line DACs will include the op-amps that were determined to yield the best performance. Often, lower end op-amps are used in evaluation boards for less expensive DACs in the vendor's line. This is more consistent with the op-amps that will be used in the final product using the less expensive DAC. Typically, the NE5532 is used for all the vendor's evaluation boards except for the top end products. AKM uses the minimum price New Japan Radio NJM 4580 on the evaluation board it sells for it entry level DACs, and the expensive TI LME49710 for AKMs newest top of the line part.
The high-performance data converter manufacturer ESS provides measurements not only for the best available operational amplifiers, but also lower priced parts.
ESS reports performance of the DAC with the following op-amp in descending performance order (page 2).
Analog Devices AD797
National Semiconductor (now a TI subsidiary) LME49710, LM4562
Texas Instruments NE5534
Similar guidelines can be found in Douglas Self's Small Signal Audio Design text, published in 2010 by Focal Press.
ESS only recommends use of the minimal cost Rohm 4560 op-amp for use with its lowest priced DAC the ES9006.
The ESS op-amp ladder has a steep cost curve. Better chips than the NE5532 may be had at a premium at about five times the cost. The Analog Devices AD797, for example, is ten times the price of a NE5332. I have never seen it in a multi-channel product.
AVRs with the LSI AVR normally deploy the lower-cost op-amps for the anti-alias and reconstruction filtering functions, though a couple use the NE5532. Pre/Pros built with SSI chips use the NE5332 or better performing op-amps exclusively. The note section in the chart in Part I identifies products with this op-amp.
Op-amps on the next rung of the ladder may be found in some SSI-based Pre/Pros, but only in segments of the signal path where the designer anticipates maximum benefit to performance. They are too expensive to be used exclusively in a multi-channel product. Other op-amps that are more expensive than the NE5532 may be utilized by some engineers, but these are not the units on the next rung of the ladder. Doug Self finds the op-amps between the rungs do not provide a significant advantage.
In some sections of an SSI-enabled Pre/Pro, a very high input impedance of the operational amplifier may be desirable. This requires an op-amp with FETs in the differential pair. No DC bias current can be present in an op-amp with a FET input. Selection of the op-amps for this application is best left to another article. All the operational amplifiers discussed above have bipolar inputs.
Power supply regulation is very important for optimal DAC performance. The ESS application note shows how different op-amp buffers for the analog power supply can make a difference to the THD and SNR of the parts. The application note points out the ESS analog output stage is very sensitive to the op-amp buffer. This is a result of how the output stage works in comparison to other DAC topologies. This is not an issue as long as the AVR or Pre/Pro designer uses the required parts and not low-cost substitutions. Yamaha reduces the output impedance in this stage by adding a transistor to the NE5332, and preceding it with an independent Ricoh R1172 voltage regulator.
The AKM evaluation board documentation sited above shows a complex local regulator which uses more than 25 parts. Such a complex local regulator would not be found in a Pre/Pro but perhaps an expensive two-channel product.
Evaluation boards for Analog Devices, Cirrus, and Wolfson use high performance single chip IC regulators. I don't give the part numbers or attempt to rank the single chip regulators in this report. Consistent with the use of minimal cost op-amps in AVRs, the voltage regulators will also often be of lower quality then what the DAC designers would like used for optimal performance.
If a cheap op-amp is used for current to voltage converter function when a current-mode DAC is specified, the performance inherent in the DAC will likely be lost. In my survey of AVR service manuals, I found companies that use a current mode DAC from TI or ADI have an NE5532 or higher-grade op-amp for the current-to-voltage converter. Some units then revisit op-amps of lesser quality for other sections of the circuitry. The current-to-voltage converter is discussed in more detail in the next section.
Unfortunately, ESS has no control of what the vendors actually use. Board photos of the new Pioneer AVR (SC-LX57), using the ESS ES 9016, show a bottom grade NJM 4570 op-amp used for the I/V and the balanced-to-single-ended converter.
Differential and Current Mode DACs
In this section, block diagrams for the analog electronics after the DACs are shown. The first diagram is for a balanced voltage mode DAC which has two outputs that are in anti-phase to each other.
The figure above shows how the output of differential voltage mode DAC is converted to a single ended signal using a balanced to single ended converter.
Distortion is reduced for two reasons:
- Subtracting these signals yields a single ended signal of twice the magnitude providing increased dynamic range.
- Even order distortion components in the individual signal, if correlated, will be canceled when the signals are subtracted.
The balanced-to-single-ended converter block is the same as the one shown in a previous figure of a balanced volume control.
In addition to the op-amp and resistors, additional reactive components may be added to form the Low Pass Filter (LPF) for the removal of high frequency, out off band energy, which occurs in the digital sampling process. More on this subject is discussed below.
Another method to reduce distortion at the output of the DAC is to transmit the signal in the current domain as shown in the figure below. In the chart from the previous part of this article, it can be seen the best current-mode DACs are superior to voltage-mode DACs for minimizing distortion.
In the current domain, the signal out of the DAC does not move in amplitude.
The green circuit block between the DAC and the balanced-to-single-ended converter converts the current flowing at the DACs output pins to a voltage. This is a current-to-voltage converter, often abbreviated as I/V converter. The circuit consists of an op-amp and a resistor in the feedback loop
Internal op-amps are absent in the typical current-mode DAC; instead, only switched current sources are present because it is difficult to integrate a high performing op-amp in the CMOS process technology.
Moving the op-amp outside the DAC permits an IC process technology designed for analog circuits to be used. The analog process technology provides improved noise performance and reduced distortion. The ability of the analog process to safely operate with a 30V power supply, rather than the 5 volts of power for the DAC, improves the signal-to-noise ratio.
Additional reactive components may be found as part of the I/V stage. These components create a low-pass filter for removal of out-of-band energy. This is typically a first-order (6 dB/octave) filter. The balanced-to-single-ended converter can provide a second-order (12 dB/octave) filtering function. By placing the poles of both filters optimally, a third-order (18 dB/octave) filter Butterworth filter is in the signal path.
Multiple DACs Combined to Produce a Single Channel Output (Mono Mode)
Multiple DACs assigned to one channel can improve the signal-to-noise ratio and, in some cases, reduce distortion. The concept was discussed in Part I of this article, but I am presenting the block diagram for the first time. Mono mode is typically used only for DACs at the top of the IC vendor's line.
Connection for a current mode DAC in mono mode simply involves attaching the leads at the output of the DACs together. This configuration allows the currents to add together and flow into the current-to-voltage (I/V) converter.
Connecting all the DACs in a mono mode configuration doubles the number of DACs for the complete Pre/Pro. Some multi-channel products use the mono configuration for the left and right channels only.
ESS DACs contain up to eight single DACs in one chip. It is possible to expand the concept by tying four of the DAC's current outputs together to form a stereo DAC or even tying all eight together to form a mono DAC.
All the application information I looked at for TI DACs with current mode outputs, on the TI website, showed circuits in which each current output of four individual DACs were converted to a voltage before the summation process to mono. The current outputs were never directly shorted. This requires two added I/V converter stages. The topology is applicable to any current-mode DAC. The TI evaluation board documentation below shows the technique:
Accuphase developed an interesting circuit to operate with the ESS octal DACs extending the concept shown in the TI application note. For stereo, there are four DACs per channel, but they do not sum all inverting (or non-inverting) current outputs together. Instead, Accuphase connects pairs of current outputs together. Two I/V converters are needed to convert the current output pairs (four I/V converters in total for the balanced signals). According to Accuphase, "A combination of current summing and voltage summing is used, for optimized operation."
Accuphase does not produce multi-channel products. It is unclear if the complex topology Accuphase uses will be used in multi-channel products using ESS parts. Since the performance improvement is likely small, this circuit would be used only with the top of the line ESS DAC.
Stereo DACs may have a digital pin that puts the DAC in mono mode. Often, the LPCM digital data to the second DAC is inverted when the DAC is in mono mode.
The current pins are configured so the DAC with the inverted data has the pins of opposite polarity connected. This ensures the current levels are doubled when the extra digital inverters are introduced into the digital LPCM stream.
Assuming correlated distortion exists between the two channels, the process of inverting the LPCM data and then connecting the current pins as shown in the figure above, may cancel correlated distortion. This is only feasible with closely-matched DACs in the same package. Data sheets I have examined do not indicate the distortion is reduced in mono mode when connected as shown above, suggesting the reduction in distortion, if any, is small.
The connection is free to the designer, since the digital inverter and its activation are done internal to the DAC chip. No reason exists not to use the configuration even if the distortion improvement is small.
DACs with the internal digital inverters do report the expected improvement in signal-to-noise ratio. The SNR is unchanged regardless of which current summation topologies is selected to create the mono DAC.
A voltage-mode DAC can be configured for mono mode. It is not possible to short the pins together as with the current-mode DACs. Since a voltage mode DAC has a low output impedance, connecting voltage outputs is similar to shorting them. The same issue occurs when one tries to parallel a pair of batteries.
For the voltage-mode output mono DACs, the balanced-to-single-ended converter must be modified so that it will perform the summation of the two in-phase voltage signals from the DAC and the two out-of-phase voltage signals from the DAC. In its simplest form, this added task is assumed by the same op-amp that performs the balanced-to-single-ended conversion. Since this change is at the circuit level of the balanced-to-single-ended converter, I am not showing a figure. The schematic of the Wolfson evaluation board documentation shows the complete circuit of a voltage output DAC in mono mode:
Like all voltage mode DACs, no I/V converters are required in the mono mode configuration.
A more complex topology to create a mono DAC distributes the balanced-to-single-ended conversion task among three op-amps. This topology is sometimes called double-balanced.
In a double-balanced DAC, all electronics of a stereo DAC are present, including a separate balanced-to-single-ended converter for the left and right channels. The output of the left channel single-ended converter produces the in-phase signal, and the right channel single-ended output produces the out-of-phase signal with the LPCM digital input inverted. Finally, an additional balanced-to-single-ended converter combines the outputs of the two preceding balanced-to-single-ended convertors.
An example where the double balanced approach is useful is an AVR with 13 (11.2) DACs. If only seven (5.2) channels are required, each DAC switches to mono mode except the DACs assigned to the subwoofer. The additional balanced-to-single-ended converter switches into the signal path of each DAC IC, creating five mono DACs from the ten stereo DACs. These balanced-to-single-ended converters (five in total) and the suite of balanced-to-single-ended converters already in place form the double-balanced topology.
The topology is compatible with current-mode DACs if four I/V converters are added. I have not supplied a figure for this configuration.
A single 9.1 AVR can operate as 7.1 with the left and right channels configured as double-balanced. An Integra product offers this option.
Quasi Current-Mode Interface for ESS DACs with a Single Operational Amplifier
An ESS DAC achieves its best THD performance when directly connected to a current-to-voltage conversion circuit (I/V) made with an op-amp section. The ESS DAC, which has balanced outputs, requires two op-amps for its current-to-voltage function. A third op-amp supports the balanced-to-single-ended converter. An identical three op-amp topology is required for DACs which have current outputs for each channel discussed above.
Traditional current-mode DACs embedded within an AVR cost at least $1,800 (refer to the chart in Part I). The combined cost of the high-performance DACs and two op-amps per channel cannot be absorbed in lower-priced AVR.
If the output pin of a current-mode DAC is shorted, its output current will flow through the short. Alternatively, when the output pin is an open circuit, the output, uselessly, moves to a power rail. The current-to-voltage converter represents a short circuit to the current-mode DAC. As the name implies, the output voltage is proportional to the DAC current.
A voltage-mode DAC has a small output impedance. In turn, the value of the voltage at the output pin is unaffected by a load provided above about 2 kOhm. The balanced-to-single-ended converter typically has a higher input impedance. Only one op-amp is needed to perform the balanced-to-single-ended operation. As discussed above, the best voltage-mode DACs have higher distortion than most current-mode DACs.
The ESS DACs use a different output stage topology. One way to model the output pin as a current source is to have it in series with a resistor with an approximate value of 780 ohms. If the output pin is shorted, the current flows through the current source into the short, and no current flows in the resistor. When the output pin is open, a voltage will appear. The voltage is found by Ohm's Law. If the current source has a value of 1 mA, then a value of 780 mV is at the output. The value of the resistor on the integrated circuit varies slightly with the voltage across it. This is the resistor's voltage coefficient which was discussed in the volume control section above. This change adds undesired distortion.
To reduce the distortion, ESS engineers have devised a novel single op-amp balanced-to-single-ended converter for the DAC. The design takes advantage of the internal resistor in the ESS DAC, using it as part of the converter. With this special circuitry, the voltage variation at the ESS output pin is reduced by ¼ compared to an open circuit. Less swing implies lower distortion.
ESS DAC distortion rises above that reported in the chart in Part I when the DAC is connected to a single op-amp circuit. The best performance requires three op-amps. AVR and Pre/Pro designers that want to replace a voltage-mode DAC with an ESS DAC of equivalent cost may use the single op-amp solution to maintain a constant external cost count. The high SNR of the ESS DAC and its ability to reject jitter are unchanged. The chart in part I identifies AVRs that use the Quasi Current-Mode Interface with an ESS DAC with note #3.
Yamaha uses this circuit in all of the current products with ESS DACs, including the top of the line CX-A5000 review soon to be published in Secrets. In the time domain, a 24 bit sine wave -90 dB below full scale shows low noise around the waveform. The distortion performance of the CX-A5000 is disappointing, although measuring at the output does not allow identification of the source as the circuit above or the single chip LSI AVR that follows in the CX-A5000.
Part III of this article series will be: An Introduction to Asynchronous Sample Rate Converters (ASRC) for Jitter Reduction and Introduction to Digital Filtering of an LPCM Signal and DAC digital filter performance. Some of this material is also covered in my review of the HK 990.
There will also be new material that includes building a Pre/Pro with pre-designed boards and methods to perform low distortion switching with SSI CMOS IC blocks. In addition, there will be a more detailed view of the internal signal flow in the LSI analog AVR chip, identifying the sources of sub-standard noise and distortion performance.