New Amplifier Technology Redefines Performance Benchmarks
NAD Masters Series M2 Direct Digital Amplifier
by: Greg Stidsen and Craig Bell
by: Greg Stidsen and Craig Bell NAD Masters Series Concept
The NAD brand has come to represent high value products that consistently offer musically honest performance. In 2005, NAD introduced the Masters Series concept as the evolution of NADs core values of performance, simplicity and value. Each component in the series offers performance, both measured and subjective, that is at the upper limits of todays technology. Elegant industrial design and impressive build quality are also essential to this concept. As a way to showcase its engineering prowess and to create a new class of NAD products, Masters Series has been a qualified success. Another key element of the Masters philosophy is the introduction of new technology that can later trickle down to NADs less expensive products. This is where our M2 Direct Digital Amplifier story begins. Because NAD considers performance to be its most important brand value, it was decided that an NAD branded Class D amplifier would not be introduced until it could offer equal or better sonic performance to NADs traditional Class AB designs.
NAD and Diodes Zetex Collaboration
One of the most promising designs we tried came from a British semiconductor company named Zetex. The circuit architecture was fully digital, which differed greatly from most of the Class D amplifiers in the market. Conceptually, Zetex had addressed the major weaknesses of previous designs and early listening tests were extremely encouraging. The sound was open and dynamic without the harshness or softness of the other solutions. Transients were superbly rendered and the amp sounded fast and tight, but not overly etched or dry. Zetex (now Diodes Zetex Ltd.) was looking for a development partner to bring their digital development out of the laboratory and into the marketplace. NADs strong engineering resources and worldwide reputation for performance proved to be a good match and so NAD and Zetex started to collaborate in 2005. Our mission was to push this new digital platform to the limit and make a very powerful amplifier as the first introduction of this new invention. The design brief for the M2 started to take shape.

The M2 Design Story

The first Class D amplifiers for audio were developed in the 1960s by Gordon Edge, a brilliant engineer working for famed British entrepreneur Clive Sinclair (Sinclair Radionics). But the performance of this early design was marginal. In the 1970s, Infinity Systems experimented with a more advanced design called SWAMP, but could not advance this development into a commercially viable product. Thirty years later, Class D amplifiers are now widely used in both professional and consumer products, but not without compromise. Noted for higher efficiency when compared with linear Class A and Class AB amplifiers, Class D still has not been able to achieve the same levels of performance, both measured and subjective, as the best Class AB designs. By the late 1990s the gap had narrowed considerably and NAD started experimenting with various Class D design solutions. But our investigations revealed that even when the amplifiers we measured performed well on paper, they did not sound very musical. Some were etched and dry sounding while others were soft and lacked high frequency extension and detail. All of them sounded less dynamic and less fluid than our reference Class AB amplifiers.
The Advantages of a True Digital Amplifier
Our initial thought was to make a basic power amp without any controls that could stand as a reference to other amplifiers. But one of the strongest attributes of a true digital amplifier is the ability to take a digital PCM signal directly, thus avoiding multiple conversions and amplifying stages that are present in all analog designs, which tend to strip the music of detail by adding hazy layers of noise and distortion. Ever since the introduction of digital source material in the 1980s, a pure digital signal path was always an attractive proposition; now it could be a reality.

NAD Masters Series M2 Direct Digital Amplifier - by: Greg Stidsen and Craig Bell
Analog Audio System Digital Storage Media D-A Converter

PCM Data Stream

Output Buffer
Although the early Zetex prototypes already had very good sound quality, the measured performance was still not quite up to the best available amplifiers. The first prototypes showed performance that was typical of the best available digital amps with a measured SNR of 105dB referenced to full power. But the NAD standard is >90dB referenced to 1 watt we needed to find an additional 10 15dB of dynamic range. With 35-bit architecture it was theoretically possible to achieve this level of performance from the Zetex DDFA (Direct Digital Feedback Amplifier) circuit, and we were determined to achieve this goal. Step-by-step, as we made measurable improvements, the subjective sound also became more detailed and expansive. This was very encouraging and motivated the team to push further and further. The benchmark SNR performance of >90dB Referenced to 1W is a difficult technical challenge, but it very closely relates to the noise performance at typical listening levels and discloses the absolute magnitude of the noise, which is the most honest way of declaring the performance. This performance level is delivered by NAD analog power amplifiers, but we must consider how this power amplifier performance fits within the overall signal chain. We can now gain insight into the difference between a conventional analog approach and a direct digital amplifier and the opportunity it presents. An extremely good digital signal source, such as a high-end CD player with a very well implemented DAC stage can offer an A-weighted SNR performance of perhaps 120dB-ref Full Scale, but the signal must be routed through a pre-amplifier stage before reaching the power amplifier stage, even if an integrated amplifier is involved. High-end pre-amps can have SNR performance ranging from 95dB up to 110dB ref FS, for the best examples. For a pre-amp, actual SNR performance is very dependent on the gain setting and can easily deteriorate by as much as 10dB from the optimal. Analyzing the overall noise generated in the signal chain quickly illustrates the dominance of the pre-amp to the overall performance, giving an endto-end performance slightly worse than this link in the chain. The example below shows how the noise is developed within the system.

Pre-Amplifier

Input Buffer Cable

Power Amplifier

Input Buffer Pre Driver Stage Driver Stage Output Stage

Tone Amp

Cable Speaker
Direct Digital Audio System M2 Amplifier

Digital Storage Media

Output Stage

Source DAC

SNR 115dB

Pre - Amp

SNR 108dB

Generated Noise 1.78uV

Generated Noise 3.98uV
The only source of noise or distortion in the M2 architecture is the final modulation and gain stage, where PCM data is converted to PWM (Pulse Width Modulation) to create the switching output. The system SNR performance is completely unaffected by the volume control setting, so audible noise is never evident at the speaker. More importantly, such a low level of noise and distortion translates directly to subjective performance. SNR performance of the M2 is 91dB ref 1W (un-weighted), in all circumstances.

How it Works

Power - Amp
SNR 115dB Generated Noise 22.5uV Amplified x40 174.4uV

175.8uV SNR = 107.1dB

These numbers are expressed in un-weighted terms.
This 107.1dB figure is referenced to a 200W output power. Stating it referenced to 1W, gives a figure of 84.1dB. This is the true end-to-end performance of even this exceptional conventional signal chain. Even the outstanding power amplifier used in this example cannot help, if noise is presented at its input. For the M2 digital amplifier, the situation is quite different. Starting at the source, a well executed digital output has no noise or distortion, beyond the fundamental limits of the number format. A Zero output is simply represented by zero in the PCM data. As it enters the digital amplifier, all pre-processing such as filters and volume controls are implemented in DSP, and as a consequence, are not subject to the effects of noise in analog components, nor tolerances in their values. The DSP architecture selected for the M2 is very high resolution, with a minimum of 35 bits at any point in the processing datapath to ensure maximum retention of resolution. Take the volume control as an example. The 35-bit system datapath means that even a 24-bit input signal has 11 bits of headroom, so the volume control can be set as low as -66dB without losing any of the original signal resolution. If the source was a CD player with 16-bit data, then the figure is astonishing at -114dB. While either of these levels is normally inaudible, the processing resolution involved is paramount to retaining and reproducing the subtleties of the original recording.
The modulation stage is the part of the M2 where the most critical innovations were required in order to hit the performance targets. The modulation process where PCM is converted to PWM is an inherently digital one, with the widths of the modulation pulses being quantized by the internal clock frequency, which in the case of the M2 is 108MHz. This means that pulses can be defined to a resolution of 9.2ns within a modulation period which is 1.18us long, for a maximum of 128 possible pulse widths. At first glance, this would seem to be much too small a number to represent an audio signal. However, there are many modulation cycles available within the period of even the highest audio frequency cycle, so noise shaping techniques can be applied to accurately resolve the signal. Put simply, varying pulse widths can be applied, so that the cumulative effect is to reproduce precisely the correct amplitude at the output. The actual amplification step is performed by the FET output stage, which amplifies the logic level pulses from the modulator output levels (3.3V) to high voltage pulses with amplitude of approximately +/-50V, and fed to the speaker terminals through an LC or reconstruction filter, which removes the high frequency energy and delivers the audio signal to the speaker. The output stage is where the very real challenges of switching power electronics arise.

In reality, the pulses at the bridge output differ hugely from the perfect shape we might hope for, and the deviation from perfection results in errors in the output signal. Most digital amplifiers are open loop and have no correction mechanism, so their performance falls far short of linear amplifiers. They cannot correct for the imperfections that are inevitable in power supplies, or real-life switching waveforms. Attempting to perfect the power supply or switching structures is not a practical approach, so the DDFA technology was developed to implement correction for the problems.
also monitored at the output of the LC filter, which means the system has amazingly low output impedance. This very tight or direct feedback path gave rise to the name Direct Digital Feedback Amplifier.
The Engineering Challenge
The engineering challenge of making this process happen cannot be understated. For example, the reference PWM signal serves as the systems definition of perfection, so it must be highly pure. One familiar metric to illustrate the difficulty: clock jitter to achieve the necessary 120dB dynamic range of the reference signal must be at a level of 5 picoseconds! (A picosecond is one trillionth of a second or 0.000 001). Another way to consider the M2 technology is to judge it as if it were simply a DAC. Think of the M2 as providing digital-to-analog conversion with amplifier gain attached. During the development, a lot of attention has been paid to assessing the performance in these terms to ensure the most accurate possible reproduction. For example, the linearity of the M2 compares well to the claims of the highest performance DAC chips.

Feedback Re-invented

An analog amplifier, whether linear or Class D, can use conventional negative feedback methods to compensate, but the problem is much more difficult for a true digital amplifier. The obvious method of digitizing the analog output and feeding back to subtract from the input is hampered by the large delays involved and results in an unstable system. To solve the problem, the Zetex team has developed an entirely new approach, which is best described as noise shaping error correction. Any deviation from the perfectly programmed pulse shape is regarded as an error. This could be caused by the amplitude of the pulse (power supply ripple or sag), the width of the pulse, or even the slope of the edges. Any of these factors will impact the area under the pulse (which is really how the signal amplitude is encoded).
The system operates by comparing the output PWM signal with a high purity Reference PWM signal to create an error signal, which is representative of the voltage error at the output. Integration in time provides an indication of the pulse area error, which is digitized at a conversion rate of 108MHz to pass back to the digital domain. The error information is then processed to compensate subsequent modulation cycles. The system can be considered to be constantly adapting to minimize the errors and hence deliver as true a signal as possible to the speaker. The output signal is

In fact the performance is so good the results at very low signal levels can be misleading. At the lower end of the linearity plot, the measurement of linearity error is influenced by the noise present. By performing an FFT on the amplifier output using multiple averages, the outstanding amplitude accuracy becomes clear. The error level is less than 0.1dB at -120dB

Audio Precision

+0 -10 -20 -30 -40 -50 -60 -70 -80 -90 -100 -110 -120 -130 -140 -140
Extremely low jitter is another impressive digital performance metric. Using the j-test to assess data-related jitter, the telltale side bands at 229Hz intervals from the fundamental are totally missing. Quite simply there is no jitter.
Audio Precision EV4 FFT 1kHz 03/26/09 10:06:41
+0 -10 -20 -30 -40 -50 -60 -70
D-A LINEARITY Data and Computed deviation

03/25/09 15:21:57

+2 +1.8 +1.6 +1.4 +1.2 +1 +0.8 +0.6 +0.4

d B r A

+0.2 +0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 -1.4 -1.6 -1.8 -120 -100 -80 dBFS -60 -40 -20 -2
-80 -90 -100 -110 -120 -130 -140 -150 -160 -170 8k 9k 10k 11k Hz 12k 13k 14k -180

d B r B

Sweep Trace Audio Precision

Color Blue Red

Line LINEARITY Axis Comment D-AStyle Thick Data Data and Computed deviation Solid Solid Anlr.Bandpass Anlr.Bandpass!Linearity Left Right

03/25/09 15:26:35

+2 +1.8

Sweep 1

Trace 1

Color Red

Line Style Solid

Thick 1

Data Fft.Ch.2 Ampl

Axis Right

Comment

1kHz, -1dBFS 44.1kHz

-104 -106 -108 -110 -112 -114 -116 d B r A -118 -120 -122 -124 -126 -128 -130 -132 -134 -136 -138 -140 -140 -135 -130
EV3 Power OP DAC linearity.at2c
+1.6 +1.4 +1.2 +1 +0.8 +0.6 +0.4 +0.2 +0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 -1.4 -1.6 -1.8 d B r A
Measured Performance of the Power Out 4ohm LoadsDigital Amplifier NAD M2 Direct 03/25/09 14:17:32 Audio Precision NAD M2 BTL THD+N vs

44.1 kHz -60dB

0.05 0.0475 0.045 0.0425 0.04 0.0375 0.035 0.0325 0.03 0.0275 % 0.025 0.0225 0.02 0.0175 0.015 0.0125 0.01
EV4 Power OP Dynamic Range FFT.at2c

-120 dBFS

Sweep 1 1

Trace 1 2

Line Style Solid Solid

Thick 1 1

Data Anlr.Bandpass Anlr.Bandpass

Axis Left Right

0.0075 0.005 0.0025 400m W
Sweep Trace Color Blue Yellow Line Style Solid Solid Thick Data Anlr.THD+N Ratio Anlr.THD+N Ratio Axis Left Left Comment
Zooming in further reveals useful resolution down to -135dB!
1kHz, -1dBFS 44.1kHz EV3 Power OP DAC linearity.at2c
The Familiar measurement of THD+N vs. output level shows the expected level of performance for a high-end amplifier. However, much more is revealed by looking at the M2 performance in terms of low level detail.
Digital Input 44.1kHz, 24bits 1kHz -4.8dB FSD Volume 0dB M2 BTL THD+N Plot.at2c

Zetex plc

M2 -120dB FFT Optical IP, Multiple volume settings

03/25/09 13:48:31

+0 -10 -20 -30 -40 -50 -60 -70 -80 -90 -100 -110 -120 -130 -140 -150 -160 -170 d B r B

M2 BTL crosstalk -26dB

03/25/09 15:14:45
-10 -20 -30 -40 -50 -60 d B -70 -80 -90 -100 -110 -120 -130 -140

10k Hz

500 Hz
Sweep Trace Color Blue Red Line Style Solid Solid Thick Data Anlr.Crosstalk Anlr.Crosstalk Axis Left Left

Sweep 4

Color Blue Red Black
Line Style Solid Solid Solid
Data Fft.Ch.2 Ampl Fft.Ch.2 Ampl Fft.Ch.2 Ampl

Axis Right Right Right

Comment Vol -20dB Vol -40dB Vol -60dB
The conventional test for dynamic range requires a -60dB output. In testing the M2, we have gone several steps further in also testing with -120dB. It is remarkable that this small signal level is resolvable. It is also notable that the performance remains exactly the same over a broad range of volume settings. The chart shows identical performance over three gain settings separated by 40dB.
Digital Input 44.1kHz M2 SEtst -60dB FFT.at2c
Channel Separation is greater than 90dB all the way up to 10kHz and still 85dB at 20kHz, which is 10dB better than the best analog stereo integrated amplifiers.
Digital Input 44.1kHz, 24bits 1kHz -26dB FSD Volume 0dB NAD BTLcrosstalk sweep.at2c

+0 -10 -20 -30 -40

M2 FFT 60Hz

03/25/09 15:03:03

+0 -10 -20 -30 -40 -50 -60 d B r A -70 -80 -90 -100 -110 -120 -130 -140
M2 IMD SMPTE 60Hz/7kHz 12dB ratio

03/25/09 14:54:28

-50 -60 -70 d B r A -80 -90 -100 -110 -120 -130 -140 -150 -160 -170 -Hz
Sweep Trace Color Blue Red Line Style Solid Solid Thick Data Fft.Ch.1 Ampl Fft.Ch.2 Ampl Axis Left Right Comment

-150 -160

Color Blue

Data Fft.Ch.1 Ampl

Axis Left
Digital Input 48kHz, 24bits, 60Hz at -3dB FS, 7kHz at 12dB down EV4 IMD_SMPTE 60Hz 7kHzTest.at2c
Distortion spectrum of 60Hz signal at 250 watts into 8 Ohms shows a very clean spectrum with higher order harmonics well suppressed.
SMPTE Intermodulation spectrum is very clean with all distortion products below -100dB or 0.001%.
NADs Unique Implementation of DDFA
The M2 has taken Diodes Zetexs core technology and integrated NADs many years of amplifier expertise to create a really unique product. The implementation of DDFA within the M2 uses customized logic within a Xilinx FPGA. This gives a custom platform on which to blend the DDFA technology with NADs own innovations within this unique architecture.
Selectable Digital and Analog Inputs This allows the M2 to be used without a preamp for the shortest possible signal path. Both balanced and single-ended signal sources are supported. Ultra-high Resolution Analog-to-Digital Conversion Analog inputs must be converted to the digital domain of the M2. NAD is using the very latest state-of-the-art devices which, along with an optimized circuit design, results in superb performance. Sample Rate of 48kHz, 96kHz and 192kHz can be selected. The performance of the Analog inputs is within 1dB of the Digital inputs for all measurements! Speaker Compensation Even though the DDFA corrects for the influence of the output filter, we offer further fine-tuning using digital filters to flatten response at 20kHz to within 0.5dB. Speaker impedances from 1 Ohm to greater than 8 Ohms are accommodated in 1 Ohm steps. Dual Mono Design Three separate ultra low noise Switch Mode Power Supplies are employed one for each audio channel and one for control. Digital Processor Loop This allows the insertion of a PC into the signal path to add signal processing such as crossover filters or room correction filters. Connection is via Optical SPDIF. Two Sets of Speaker Outputs This is to allow for convenient Bi-Wiring of loudspeakers. Large VFD Display The front panel display shows selected input, incoming Sample Rate, ADC Sample Rate, Volume Setting and Mute. Each input can be renamed if desired. System Remote Control IR remote controls NAD CD Players as well as the M2. Front IR Sensor and Rear IR Input. 12V Trigger Input Allows for automated On/Standby function. RS-232 Port Interfaces to third-party automated control and allows for software updates.
Digital PowerDrive The benefit of NADs PowerDrive technology is the ability to have high dynamic power necessary for accurate reproduction of musical transients coupled with the ability to drive difficult speaker loads without increased distortion. The DDFA architecture allowed us to translate this important attribute into the digital domain giving a highly accurate power/time envelope that closely matches our ideal based on NADs research of recorded music requirements. Digital Soft Clipping NAD first developed Soft Clipping in the 1970s as a way to avoid the harsh and highly non-musical sound when an amplifier is driven beyond its limits. Soft Clipping allows a graceful overload without the usual generation of high order harmonic distortion that normally occurs as the sine wave gets squared off. Now digitally controlled, it can be carefully modeled for ideal results. Perfect Digital Volume Control Signal-to-noise, distortion and channel separation are the same for all control settings. Channel tracking is perfect. There is a fixed setting available for analog inputs allowing the M2 to be used as a basic power amp with analog preamps if desired.

NAD M2 Direct Digital Amplifier
Inventing the Future of Audio

Robert Harley

he term digital is often erroneously applied to amplifiers with Class D (switching) output stages, but in the case of NADs new M2 Direct Digital Amplifier that word is appropriate. In fact, the M2 represents a major rethinking of audiosystem architecture, directly converting standard-resolution or high-res digital bitstreams into signals that can drive loudspeakers. Functionally, the M2 is an integrated amplifier that replaces a DAC, preamplifier, and power amplifier. The M2 eliminates from a traditional signal path all the electronics of a DAC as well as the active analog gain stages of a preamplifier and power amplifier. It does this by converting the PCM signal from a digital source directly into a pulse-width modulation (PWM) signal that turns the M2s output transistors on and off. Thats itno digital filter, no DACs, no multiple stages of analog amplification, no interconnects, no jacks, no analog volume control, no preamp. The conversion from the digital domain to the analog domain occurs as a by-product of the switching output stage and its analog filter. This is as direct a signal path as one could envision. (See sidebars for the technical details.) NADs M2 is a significant departure for the company that made its reputation building simple and affordable electronics. For starters, the M2 costs $5999, a new price level for a NAD integrated amplifier. Second, the M2 is NADs first amplifier to use a switching output stage. The company had previously rejected the technology in favor of linear amplifiers because switching output stages just didnt sound good. But the M2s output stage is significantly different from any other currently offered (see sidebar). Third, NAD believes that the M2s technology could eventually become the basis for nearly all of its amplification products. In fact, NAD suggested that the M2 was not designed to capitalize on Class Ds functional advantages, but rather to establish a new benchmark of performance in amplification, no matter what the technology. Lets look at the M2 Direct Digital Amplifier in operation. The unit looks and functions like one of NADs upscale Masters Series integrated amplifiers, with a row of front-panel input-select buttons, a volume control, and a display. The rear panel, however, reveals that the M2 is not a conventional integrated amplifier. Five digital inputs are provided (two RCA, one AES/EBU, two TosLink, plus a TosLink loop) along with one single-ended and one balanced analog input. The digital inputs can accept any sampling
136 December 2009 The Absolute Sound
frequency from 32kHz to 192kHz. Analog signals fed to the M2s analog-input jacks are converted to digital. Once youve connected an analog or digital source to the M2 (such as a CD transport or music server) and loudspeakers via the output binding posts, the M2 functions just like a traditional integrated amplifier. You select the source from the front panel and control the volume with the large front-panel knob or from the remote control. The front-panel display shows the input sampling frequency and volume setting. Purists will note that the M2 requires that analog signals, such as a phonostage output, be converted to PCM digital. Similarly, those who enjoy SACD will be loath to convert their SACD players analog output to PCM, and then back to analog in the M2. The M2 offers a number of features not found on a traditional integrated amplifier. Pushing the Menu button allows you to select the sampling frequency of the analog-to-digital converter (for analog input signals) as well as engage an upsampling feature that converts, for example, 44.1kHz to 96kHz. Analog signals are digitized at up to 192kHz/24-bit. You can also attenuate the level of the analog inputs by up to 9dB. A Speaker Compensation adjustment is a five-position adjustment that allows fine tuning of the top octave to match the speaker impedance. An absolutepolarity switch rounds out the menu-accessible features. A rearpanel switch engages NADs Soft Clipping feature, which limits the output to prevent audible distortion if the amplifier is overdriven. An RS232 port allows external control via a PC or control system such as Crestron or AMX. The full-function remote control selects between sources, adjusts the volume, dims the display, and can also control a NAD CD or DVD player. The M2 doesnt seem like a switching amplifier in operation; it is heavier than most Class D amps and although it runs cooler than a traditional Class AB amplifier of comparable output power, it produces more heat than any other Class D amplifier Ive had in my home. Listening I lived with the M2 for a couple of months, driving the Wilson Audio Alexandria X-2 Series 2 loudspeakers as well as the YG Acoustics Kipod Studio (review forthcoming). When driving the Kipod, the M2 could drive only the upper module, not the powered woofer that accepts a line-level input. I also heard the M2 with the

The Cutting Edge

Technology: Not Just Another Switching Amplifier
The M2 is different in two important ways from other amplifiers that use a Class D switching output stage. In a conventional switching amplifier, analog input signals are converted to a series of pulses that turn the output transistors fully on or fully off. The signals amplitude is contained in the pulse widths (see sidebar PulseWidth Modulation). An output filter smooths the pulses into a continuous waveform. But in the M2, PCM digital signals fed to the amplifiers input (from a CD transport, music server, or other source) stay in the digital domain and are converted by digital-signal processing (DSP) to the pulsewidth modulated signal that drives the output transistors. This difference might not seem that great at first glance, but consider the signal path of a conventional digital-playback chain driving a switching power amplifier. In your CD player, data read from the disc go through a digital filter and are converted to analog with a DAC; the DACs current output is converted to a voltage with a current-to-voltage converter; the signal is low-pass filtered and then amplified/ buffered in the CD players analog-output stage. This analog output signal travels down interconnects to a preamplifier with its several stages of amplification, volume control, and output buffer. The preamps output then travels down another pair of interconnects to the power amplifier, which typically employs an input stage, a driver stage, and the switching output stage. In addition to the D/A conversion, thats typically six or seven active amplification stages before the signal gets to the power amplifiers output stage. To reiterate the contrast with the M2, PCM data are converted by DSP into the pulse-width modulation signal that drives the output transistors. Thats it. There are no analog gain stages between the PCM data and your loudspeakers. The signal stays in the digital domain until the switching output stage, which, by its nature, acts as a digital-to-analog converter in concert with the output filter. The volume is adjusted in DSP. The second point of departure between the M2 and all
138 December 2009 The Absolute Sound
other Class D amplifiers is the switching output stage itself. NAD partnered with the U.K. design team of the American semiconductor company Diodes Zetex, who had developed a novel switching-amplifier technology. NAD engineers worked with Diodes Zetex for more than four years to improve upon Zetexs basic idea before it was ready for the M2. Diodes Zetex calls its amplifier a direct digital feedback amplifier (DDFA). The primary innovation is the use of feedback around the output stage to reduce distortion. Feedback, used in virtually all linear amplifiers, takes part of the output signal, inverts it, and sends it back to the input. The technique lowers distortion. But feedback isnt practical in switching amplifiers because of the delay involved in sending part of the output signal back to the input. Switching stages operate on extraordinarily precise timing; a glitch of a nanosecond can cause the output stage to lock up. The Zetex innovation is to compare the actual high-level PWM signal (at the transistor outputs) to a lowlevel reference PWM signal. Any difference between the actual and reference PWM signals represents a voltage error. The actual PWM signal can deviate from the theoretical ideal because of power-supply noise or droop (a drop in

Volent Paragon VL-2, a $5000 stand-mounted two-way employing a Heil Air-Motion Transformer (also on-deck for review). I compared the M2 to my usual system of a Berkeley Audio Design Alpha DAC, Pass Labs XP20 preamp, and Pass Labs XA100.5 Class A power amplifier, all connected with MIT MA-X interconnect and MIT Oracle MA loudspeaker cable. Note that the M2 functionally replaces this entire Berkeley DAC/Pass preamp/Pass power amp/MIT system, and costs about onetenth the price. The digital source for both systems was the AES/EBU output from a Class Audio CDP-502 to play CDs. I tested the M2 with high-resolution bitstreams sourced from the fan-less, drive-less, PC-based music server built by Boston retailer Goodwins High-End and described in Issue 189. When I connected the AES/EBU output from the server into the M2s AES/EBU input, the M2 instantly locked to any sampling frequency and was glitch-free. I experienced two minor operating problems with the M2. First, the protection circuit triggered a couple of times, even with no music playing. Turning off the power reset the circuit. Second, when I turned on the M2 on one occasion I heard noise from the right channel. Turning off the unit and turning it back on corrected the problem. This happened only once in dozens and dozens of power-up cycles. Long-time readers will know that Im no fan of switching amplifiers. They have their virtuessmall size, very little heat dissipation, light weight, and usually a considerable amount of output power for the money. But when the music starts, Class D amplifiers have left me cold. They can sound very dynamic, but exhibit considerable variability in sound quality depending on the loudspeaker they are driving, the cables, and other factors. The switching amplifiers Ive heard (admittedly, I have not heard many) have exhibited a mechanical character, along with a chalky coloration in the midrange that robs instruments of their distinctive tone colors. But the M2 sounded completely unlike any other Class D amplifier Ive heard. It had no characteristic fingerprint that identified its technology. Rather, the M2 tended to get out of the way, reflecting the virtues and verities of the recording. Unlike other switching amplifiers Ive heard, the M2s departures from neutrality were subtractive rather than additive. That is, it commits sins of omission rather than sins of commission. The M2 sounded like a very high-quality conventional (linear-amplification) playback system in many ways, with one notable exception; this amplifier was dead-quiet at any listening level and with any loudspeakereven the 95dB-sensitive Wilson X-2. Backgrounds were truly and totally black, a quality that gave instrumental images a greater tangibility, both spatially and texturally. The dead-silent background seemed to throw instrumental images into sharper relief, enhancing the impression of three-dimensional objects existing in space. This palpability was also partially the result of the M2s somewhat forward spatial perspective which puts the listener around Row E. The M2 also tended to spotlight the midrange to some degree, again adding to the impression of presence and the palpability of instrumental and vocal images. This was generally an appealing quality, although some forward-sounding and midrange-emphasized recordings, such as In Other Words from The Teodross Avery Quartet, were not complimentary to the M2. Conversely, naturally recorded vocals such as the outstanding ReVisions: Songs of Stevie Wonder by Jen Chapin, took on a you are there quality that was extremely involving. The M2s bass was simply greatextended, rich, warm, powerful, and muscular. The bottom end was rich and densely saturated in tone color, wonderfully nuanced and articulate, and very fast and dynamic. I greatly enjoyed the M2s combination of weight and agility on acoustic and electric bass, particularly with virtuoso playersStanley Clarkes acoustic bass on The Rite of Strings with Al DiMiola and Jean-Luc Ponty, for example. Left-hand piano

The Absolute Sound December 2009 139
voltage), slight changes in the pulse widths, transistor tolerances, or variations in the rise-time of the pulse edges. All these potential sources of errors affect the area under the pulses, which is how the analog amplitude is encoded. This error shows up as a voltage, which is digitized at a conversion rate of 108MHz, processed to compensate for subsequent modulation cycles, and then fed into a noise-shaper that adjusts the pulse shape, on a continuous basis, to compensate for errors in the output stage. In addition to decreasing distortion, this technique also lowers the amplifiers output impedance. The reference PWM signal must be essentially perfect or else the system will correct errors that arent present. The pulse widths must be precise to within five picoseconds, a level of performance commensurate with the lowest clock jitter in state-of-the-art digital-to-analog converters. In fact, you can think of the M2 as a DAC with gain and judge its technical performance using the same metrics as those employed in evaluating D/A quality. For example, at -120dB, the M2s linearity error is less than +/-0.1dB (an amazing spec, by the way), and the unit provides useful resolution down to an astounding 135dB. The M2s topology has interesting ramifications for a systems overall noise performance. In a traditional system of digital source, analog preamplifier, and analog power amplifier, any noise introduced ahead of the power amplifier greatly degrades the systems signal-to-noise ratio (SNR). For example, if we start with a CD player with a SNR of 115dB, feed its output to a preamplifier with a SNR of 108dB, and then drive a power amplifier whose intrinsic SNR is 115dB (all great specs), the systems overall SNR is only 84.1dB referenced to 1W (all SNR numbers are unweighted). Noise at the front of the chain gets amplified by the power amplifier, no matter how quiet that amplifier is. In the M2, the only source of noise is in the DSP and the switching output stage, and the noise level is completely independent of the gain. That is, the SNR doesnt degrade at low volume. The DSPs noise is kept low in part because of the 35bit data path. The M2 has an SNR of 91dB (unweighted, referenced to 1W) at any signal level. Indeed, I turned the gain all the way up and put my ear next to the tweeter of the highly sensitive Wilson Audio Alexandria X-2 Series 2 loudspeaker (95dB 1W/1m) and heard no noise. Theres no free lunch, however. Switching amplifiers require a serious output filter (typically a large inductor and a capacitor) to remove high-frequency switching noise from the output, and to smooth the waveform. This filter is conceptually similar to the reconstruction filter in traditional digital-to-analog conversion. Switching amplifiers are also very susceptible to audible degradation if the power supply feeding the output transistors isnt perfectly clean. Thats because the output transistors either connect the output transistors powersupply rail to the loudspeaker (in the on state) or disconnect them (in the off state). Any noise or ripple on the supply rails is connected directly to the loudspeaker. Switching amplifiers thus require an extremely quiet supply. Nonetheless, many switching amplifiers skimp on the power supply in an effort to keep size, weight, and cost low. The M2 has a more substantial power supply than Ive seen in any other amplifier with a switching output stage. Three separate supplies are used, one for each audio channel and one for the control circuitry and housekeeping. Each of the M2s amplifiers is contained on a roughly 6"-square circuit board and heat-sink assembly that attaches to a mother-board below it. It appears that each channel employs two pairs of output transistors. The rear panel is shielded, presumably to prevent radiated switching noise to get into the signal after it has been filtered. The chassis is segmented into two additional shielded modules, again to protect against switching noise pollution generated by the output stage.

140 December 2009 The Absolute Sound
lines were also well served by the M2s dynamic agility and powerful bottomend (the Beethoven Piano Concertos led by Sir Colin Davis on the Pentatone label). The M2 conveyed the impression that it took tight-fisted control over the woofers, backed up by tremendous energy reserves. The articulation in the midbass was extraordinary; I could easily hear the initial transient of plucked acoustic bass strings, followed by the rich resonance of the instruments body. When an audio product performs in many ways above its price class as the M2 does, theres a tendency to judge all areas of performance against its strengths. In other words, the product itself raises its own performance bar. Keeping that in mind, I noticed a trace of hardness in the upper midrange that manifested itself as a glare on certain instruments, particularly the upper range of trumpet. This is a common characteristic of amplifiers of this price, but it was different in the M2. Where most amplifiers impose this characteristic over a wide band that makes itself nearly always audible, the M2s coloration was confined to a relatively narrow band. Consequently, I heard it only occasionally when there was energy in that region. This slight coloration didnt bother me during extended listening to the M2 alone, but was apparent when I compared it to my reference system of the Berkeley Alpha DAC and Pass XA100.5 pure Class A power amplifiers. The M2 didnt have quite the timbral liquidity and midrange warmth of the reference system. Nonetheless, the M2s overall sound was smooth and relaxed. The treble tended to favor ease over the last measure of detail. The top octave wasnt quite as open, extended, or transparent as my reference system. Listening to a straight-ahead jazz CD I had engineered live to two-track (Confirmation by the Chiz Harris Quartet), drummer Harris cymbals were not quite as vibrant. Similarly, Conte Candolis flugelhorn took on slightly more of a golden and burnished hue than it had in life. If a component departs from neutrality, its better that this departure be in the direction of slightly softening of the treble rather than emphasizing it. I should reiterate that you can adjust the M2s treble
balance to match your system via the front-panel menu. The M2 sounded quite detailed, although the very finest inner detail was not as nuanced as that heard in the reference system. The M2 didnt resolve the last measure of information that conveys the mechanism by which a sound was created. For example, theres a passage in Sorceress from Return to Forevers Romantic Warrior (on the newly remastered The Anthology CD) in which Lenny White overdubs an intricate percussion figure on timbales in counterpoint to his drumming. The reference system better revealed the nature of the timbales, making them sound more like instruments being struck and less like mere transients. The M2 was outstanding in its ability to unravel complex musical lines. Many amplifiers of this price tend to have a flat homogeneity that prevents one from hearing quieter instrumental lines in the presence of louder ones. This aspect of music reproduction is crucial to understanding the intent of the composer or performers. The M2 was the antithesis of smeared, congested, or confused. Instead, it laid out with exquisite resolution everything that was happening in the music. Moreover, it did this in a completely natural and organic way, with no trace of the analytical. Partly as a result of this quality, and partly a result of the M2s fabulous way with dynamic contrasts and shadings, music always had an energetic and upbeat quality. I could feel the spontaneous music-making on the previously mentioned Confirmation disc Id engineered and remembered from the session. The M2 had a rhythmic coherence and sense of life that thrilled me and riveted my attention on the music. Interestingly, I noticed this quality most on bebop; Freddie Hubbards solo on his great composition Birdlike from pianist George Cables Cables Vision positively soared. Finally, the M2s A/D converter (fed by the Aesthetix Rhea Signature phonostage) was very good, but not completely transparent. It shaved off a bit of resolution at lowest levels and very slightly hardened timbres.

Pulse-Width Modulation

How can a series of pulses represent the continuous waveform of music? In exactly the same way that Direct Stream Digital (DSD), the encoding format behind SACD, produces music from a bitstream. In fact, PWM and DSD are conceptually identical. Fig.1 shows the relationship between a DSD bitstream and the analog waveform that bitstream represents. The bitstream is a series of pulses of varying lengths, with the pulse length encoding the analog signals amplitude. The pulse-train generated by DSD encoding looks remarkably analog-like. That is, you can look at the pulse train and get an idea of what the analog waveform looks like. The relationship between the analog signal and the bitstream is so close that in theory, a DSD signal can be converted to analog with a single capacitor (DSD-to-analog conversion is more complex in practice). The bit rate of DSD as used in SACD is 2.8224 million bits per second.
Encoding is sometimes called Pulse Width Modulation (PWM).
1010110111011110111110111101110110101010101010010001000010000100010010101010
Pulse-Width Modulation represents an analog waveform with a series of varying-length pulses.

The M2s Predecessor

The M2 isnt the first switching amplifier to convert PCM to PWM. That distinction belongs to the TacT Millennium, which I reviewed at its introduction in 1999. But the M2 isnt simply a more modern version of that topology. Rather, the M2 employs an entirely new and radically different switching output stage (see sidebar). In addition, the Millennium adjusted the volume by changing the voltage of the power supply rails feeding the switching output transistors. The M2 adjusts the volume in the digital domain with the same digital signal processing (DSP) chip that performs the PCM-to-PWM conversion.
142 December 2009 The Absolute Sound
In a switching amplifier, the output transistors are turned fully on or fully off by the pulse-width modulated signal. The analog signals amplitude is encoded as the area under the pulses; longer pulses (longer on times for the output transistors) represent a higher analog-signal amplitude. This is contrasted with traditional linear amplifiers in which the output transistors are in a continuously variable state of conduction. The output of the PWM stage is a series of high-level pulses that must be smoothed into a continuous waveform. Every amplifier with a switching output stage employs a large filter (an inductor and a capacitor) between the output transistors and loudspeaker terminals to perform this smoothing function and to remove switching noise. In the Diodes Zetex amplifier module, the pulses are quantized at 108MHz. This frequency determines the number of discrete pulse widths available to represent the audio waveform. That number is 128, which appears at first glance to be too low to encode a complex musical signal. But even at 20kHz, there are many modulation cycles available within the period of a 20kHz waveform.

Cutting Edge

Conclusion Despite costing one-tenth as much as my reference system (all the components of which are outstanding), the M2 was extremely engaging musically. Overall, I preferred the reference system, but not by as much as the price disparity would suggest. I usually wouldnt judge a $6000 product against one costing more than $50k, but the M2s outstanding performance in many areas invited the comparison. Moreover, the M2 represents a radically different approach to amplifier design, digital-to-analog conversion, and system architecture. As such, I evaluated how the M2 sounds not just in comparison with similarly priced conventional amplification and digital-to-analog conversion, but how its new technology stacks up on an absolute basis. (You should consider this when reading how the M2 falls short of a reference-quality system. I included those observations not to diminish the great achievement the M2 represents, but to put this new technology in context.) As for the M2 as an alternative to a $3500 conventional integrated amplifier and a $2500 digital-to-analog converter, its a slam dunk. I havent heard, nor can I imagine, any combination of amplification and DAC at the price approaching the M2s performance. Moreover, the M2 delivers, in one chassis, decoding of high-resolution digital audio, the source-switching and control functions of a preamplifier, and 250W of amplificationall with outstanding ergonomics. I can envision the M2, or its descendents, as part of a three-piece playback-system: music server, M2-like product, and loudspeakers. NADs M2 is a triumph on many levels, not the least of which is that it points toward a new direction in amplifier design and system architecture. I predict that years from now audiophiles will look back on the M2 as the progenitor of the next generation of audio.
STATEMENT OF OWNERSHIP, MANAGEMENT, AND CIRCULATION (39 USC 3685) 1. Publication Title: The Absolute Sound 2. Publication No.: 0494150 3. Filing Date: 10-1-09 4. ssue Frequency: Jan, Feb, Mar, Apr/May, June/Jul, Aug, Sept, I Oct, Nov, Dec 5. No. of Issues Published Annually: 10 6. nnual Subscription Price: $36.00 A 7. Mailing Address of Known Office of Publication: NextScreen LLC, 4544 S. Lamar #G300, Austin, TX 78745. 8. Mailing Address of Headquarters or General Business Office of Publisher: NextScreen LLC, 4544 S. Lamar #G300, Austin, TX 78745 9. Names and Mailing Addresses of Publisher, Editor-In-Chief, and Managing Editor: Publisher: Jim Hannon, Palo Alto, CA 94303. Editor-In-Chief: Robert Harley, Tijeras, NM 87059 Executive Editor: Jonathan Valin, Cincinnati, OH 45202 10. Owner: NextScreen LLC, Inc., 4544 S. Lamar #G300, Austin, TX 78745. Thomas B. Martin, 1012 Weston Rd., Austin, TX 78746 Harry Pearson, Box 235 Seacliff, NY 11579 11. Known Bondholders, Mortgagees, and Other Security Holders Owning or Holding 1 Percent or More of Total Amount of Bonds, Mortgages, or Other Securities: None 12. Tax Status: Has not changed during preceding 12 months 13. Publication Title: The Absolute Sound 14. Issue Date for Circulation Data Below: Oct 09 15. Extent and Nature of Circulation:

Average No. Copies Each Issue During Preceding 12 Months
No. Copies of Single Issue Published Nearest to Filing Date
Comment on this article on the Forum at avguide.com

SPECS & PRICING

Continuous output power: 250Wpc (8 ohms); 250Wpc (4 ohms); 300Wpc (2 Ohms) IHF dynamic power: 300W (8 ohms); 450W (4 ohms); 600W (2 Ohms) Peak output current: >60A Signal-to-noise ratio: >120dB (Aweighted, referenced to 200W) Digital inputs: S/PDIF on RCA jacks (x2), AES/EBU (x1), TosLink optical (x2) plus TosLink in/out loop Sampling frequencies supported: 32kHz192kHz up to 24 bits Analog inputs: Unbalanced on RCA jacks, balanced on XLR jacks Analog-to-digital converter: Fully balanced, 192kHz/24-bit Dimensions: 17.12" x 5.24" x 17.87" Weight: 44.45 lbs. Price: $5999 NAD Electronics Intl 633 Granite Court Pickering, Ontario, Canada L1W 3K1 (905) 831-6555 nadelectronics.com
a. Total no. Copies (Net Press Run): b. Paid and/or Requested Circulation: (1) Paid/Requested Outside-County

44,685

36,035
21,(3) Sales Through Dealers & Carriers: 11,196 (4) Other Classes Mailed Through the USPS 0
Mail Subscriptions: (2) Paid In-County Subscriptions: c. Total paid and/or Requested Circulation (Sum of b (1), (2), (3), and (4)): d. Free Distribution by Mail:
0,0 7,27,28,127 7,908 36,035 97.8%
33,e. Free Distribution Outside the Mail: 700 f. Total Free Distribution (Sum of d and e) 1,312 g. Total Distribution (Sum of c and f): 34,467 h. Copies not Distributed: 10,218 i. Total (Sum of g, h1, and h2): 44,685 j. Percent Paid and/or Requested Circulation: 96.2%
The Absolute Sound December 2009 145