The MU2 ‘Major DAC’

In this article we present some backgrounds of the design principles of the Grimm Audio ‘Major DAC’ digital to analog converter in our MU2 music streamer. Converting a digital signal into analog has always been a major challenge, which is clearly demonstrated by the abundance of DA conversion concepts that have seen the light over the years. In our experience high quality audio requires an extreme linearity and dynamic range. Even though modern chip technology allows for very advanced implementations, none of these seem to fully satisfy the high-end audio enthusiast. To circumvent these limitations a few companies acquired more design freedom by using a ‘discrete DAC’ approach in which all digital and analog stages of the DAC are developed in house. This clearly was the way to go for us as well. In a project of three years we created a unique digital to analog converter that has an optimal combination of dedicated DAC pre-processing in FPGA and our own discrete DAC hardware.

Existing techniques either use multi-bit conversion, a single-bit (bitstream) conversion, or a smart combination of these called Pulse Width Modulation (PWM) conversion. Conventional multi-bit conversion uses a voltage level per bit, which demands an extreme precision of the linearity of the larger bit steps. For example, the largest bit step is many thousands times larger than the smallest bit step, but should have the same precision as that smallest step, otherwise adding that small step makes no sense. To reach more than 18 bits precision with this technique is extremely difficult. As a result these types of DACs show distortion of micro-dynamics in the music. Also, cross-talk from the conversion control signals can cause graininess or harshness in their sound character.

To get around these problems, Philips introduced the ‘bitstream’ single bit technology end of the 80’s. Later it was adopted for SACD as ‘DSD’ format. In theory, a single bit type of DAC is inherently linear, as the single bit voltage level is always ‘precise’. But a single bit converter of course can only represent two signal levels instead of the thousands of a multi-bit, which means it has a much higher noise level. To make this technique work for audio, the actual conversion is created by very fast switching of the single bit value through oversampling, and then pushing the high noise away from the audio band into the inaudible region above 20 kHz by means of a noise shaper. 

This process brings other challenges to a DA converter unfortunately. For instance, there are more stringent requirements on the jitter performance of the clock. Also, this technique unavoidably puts a significant amount of high frequency energy on the DAC output, which can be tough for the following amplification stages. Moreover, in a single bit architecture, the high frequency components modulate in a way that’s depending on the audio signal. Because of this signal dependent efficiency, digital noise shaper systems operate with limited ‘stability’. As a result, the transparency and audible coherence are compromised.

In a PWM DAC, the disadvantages of the single bit architecture have been overcome by using a couple of bits (e.g. 5) and then representing their values by a varying width of the single bit stream (PWM stands for “Pulse Width Modulation”). Since there are still just two voltage levels, the system remains inherently linear. Again, a noise shaper is required to create the DAC signal. However – unlike with single bit architecture – the noise shaper now operates with a constant efficiency. On the downside an even higher clock frequency than with the single bit solution is required. And, more importantly, the implementation of such a noise shaper demands extreme processing power. In any practical realization, this results in a compromised implementation, which inevitably translates to a compromised sound quality. 

The MU2 ‘Major DAC’ has been placed on the optimal middle ground between all these options. It is using a 1.5 bit architecture. Amplitude linearity is inherently guaranteed as the 1.5 bit value is represented with a single bit D/A cell, in ‘PWM style’. Like with PWM DACs, the noise shaper is running with an effectively constant efficiency, realizing a linear operation over the entire dynamic range, as illustrated in the graphs below. 

This still requires extensive processing power, which – thanks to the lean 1.5 bit architecture – can be exploited completely to realize a flawless system in a powerful FPGA. The implemented solution in the Major DAC results in a zero error operation of the noise shaper. Moreover, the 1.5 bit DAC choice offers such a stable noise shaping operation that it allows for a highly optimized, unique 11th (!) order noise shaper. 

To filter the resulting strong high frequency noise before it enters the analog signal path a so-called FIR DAC topology is employed, using 16 DAC cells per channel. And while we are on the topic of filtering: worth noting is that the input of the noise shaper is fed from our extreme precision “Pure Nyquist” digital FPGA filter, running at 128 times the base rate. All these measures enable the Major DAC to reproduce micro-details in the audio that have never been heard before. 

Special care is taken in the analog signal path to preserve the beauty and transparency of the DAC output signals. The analog signal path is implemented fully symmetrically using very high quality circuitry, components and layout. The signal is routed through a first-class, relay based volume control section so that the analog output of the MU2 can be fed directly into a power amplifier. Additionally, up to two analog sources can be connected, allowing the MU2 to be both the digital and analog ‘hub’ in your HIFI system.

Major DAC features:

  • 11th order flawless Noise Shaping
  • 1.5bit architecture, 1bit cell, at 512fs rate
  • 128fs Pure Nyquist oversampling principles used throughout
  • Smart, extended settling technology
  • Intersample overs supported without clipping
  • 16-tap FIR-DAC topology
  • Fully symmetric analog signal path
  • Ultra stable local power supply technology

In the left column a typical 7th order 1 bit noise shaper is shown at three audio levels of 0 dB, -40 dB and -80 dB. In the right column you see the 11th order Major DAC noise shaper of the MU2. Note that in the 1 bit case the high frequency spectrum is slightly dependent on the signal strength, where the spectrum of the MU2 1.5 bit noise shaper always remains stable. This translates to a higher linearity in the audio band.
Tags: ,