"Dynamic range. The higher voltage for this model shows a very nice dynamic range of 125dB! You also can see the low noise floor due to this approach." That's nearly 21-bit resolution.
"The linearity plot shows an analog deviation of maximally +5dB at the -140dB point."
We're back at smart not bling engineering. "There are still many difficult steps to go and many auditions to determine where to set the final sonic balance." That's our cue for the necessary influence of subjectivity and taste. Whatever residual THD remains in a circuit for example, it often can be set so either the 2nd harmonic dominates or the 3rd or both remain equal. Where to set it then? That's determined by ear and with a decision. 0.00% distortion is a myth after all. What if two parts measure but don't sound the same? Again, an audio analyzer can't make that decision. What if optimizing one measurement worsens another and vice versa like an unbreakable teeter totter? These examples all exceed sheer specmanship. They involve sonic judgment calls which are made by people not machines. That too is part of an engineer's job. It must weigh personal taste against popular taste. Anything which doesn't sell direct makes shopkeepers the primary buyers. They must like it and agree to represent it before an end user ever gets it into their home. Setting the final sonic balance thus is a vital judgment call. Sales, success and business survival all depend on it. No pressure.
Segmented R2R. "We use two 16-bit ladder DACs to obtain our desired 24-bit depth. A very fast FPGA inside each module handles this bit splitting. We do it because achieving DAC linearity in the less significant bits is notoriously difficult. So we send just 12 bits to the most significant part of our ladder then do the same for the least significant bits at the second ladder. This renders the very low-level signals far too loud which we correct in the analog domain. The glue logic for the two signal halves and subsequent analog attenuation of the lower half is a tricky job but confers a clear advantage. Switching noise amplitude is always far below the analog signal and especially so for the part of the least significant bits where we apply 64dB attenuation. Now our segmented R2R processing creates straight-line signal linearity down to -140dB." That already was true for Morpheus but Father applies twice the conversion horse power or double the number of modules.
July 6th. "I'm busy with experiments focused on the DAC module. As you probably noticed, there's a global shortage of electronic parts so lead times are much higher. Any investigation of alternate parts is delayed until my orders come in."
July 12th. "It's not always disadvantageous when parts availability is as poor as it is now. So I replaced a number of parts on the mother board where each had to first be evaluated for causing no detrimental effect. Recently we became a little concerned when we could no longer source the parts we apply to the I/V section of our converter modules. These parts influence performance. To evaluate alternatives, a number of quick adjustments were necessary to shorten the redesign process. We built a test module with the sole intention of quick optimization checks. The above photo shows an R2R module with potentiometers which adjust the main parameters of the I/V section. By comparing new design to old, it was easy to identify the differences. Especially important are the low-level values of stimuli and behavior like open-loop bandwidth, noise, the common-mode rejection and power supply rejection ratios plus very low output impedance across the spectrum. If that's too high, it can affect the ladders' linearity. Long story short, we realized that with the new parts, we halved almost all our distortion figures. Those gains became particularly visible with the weaker signals. After isolating the optimal values, the final design will lock them in with fixed parts so no pots. Voilà, moving toward our latest SDA-3 converter module."
