So here is my technical presentation about Mojo 2 - the changes from Mojo.
The big change here is the removal of the coupling capacitor - this gave Mojo it's distinctive warmth. It's a little strange that just one cap can have so much difference as other DACs are full of capacitors. How can one simple passive component make so much difference? I sometimes think that tonal balance is a little like balancing scales, and as one improves transparency the weight and sensitivity of the scales become more and more sensitive - so on other DACs a small change would not be a big deal, but because of Mojo's innate transparency, very small technical changes can have a profound effect.
The next big change is the noise shaper and pulse array DAC, which accounts for most of the improvements in depth perception.
But with Mojo 2 having a more neutral tonal balance, how can one tune the sound quality to taste without lossy DSP?
The key here is being able to define what perfect means for EQ; and then being able to measure the performance - if you can't measure the actual performance, then "perfect" "transparent" or "lossless" are mere words with no verifiable meaning.
There are three important features to transparent performance - small signal amplitude accuracy, small signal phase accuracy and finally noise floor modulation.
Small signal amplitude accuracy is something I have known about for many years, with very many listening tests confirming it's importance. Whilst developing Dave, I could put a number on the requirement - basically, in order to preserve the perception of depth, I needed to be able to reproduce a -301 dB signal with an amplitude accuracy of +/- 0.001dB - these numbers are confirmed through digital simulation - in short digital domain measurements. If a noise shaper or digital module failed this test, it would degrade depth and detail perception. So if we want to preserve depth (lossless depth perception) its essential that any DSP needs to pass this demanding test.
Small signal with amplitude phase accuracy is when you compare the phase shift from a 0dBFS signal, against the same signal but at -301dB. Again, to pass the test the phase shift needs to be identical with the different amplitudes (identical meaning within +/- 0.001 deg). This realisation that digital circuits phase shifts are non-linear with amplitude started with work onto why the Hugo M scaler had better perception of depth. Since my WTA filters are small signal amplitude perfect, it implied that phase shifts with amplitude could degrade depth perception too. I tested this out with work upon Mojo's improved noise shapers - the original Mojo noise shaper did have a phase error at -301dB, and eliminating this gave much better depth perception.
The final issue is noise floor modulation, where the noise level changes with signal level. Noise floor modulation is very audible; I have heard effects when the noise floor modulation is well below measurable limits. Eliminating noise floor modulation in digital modules is possible with fixed point architecture and using noise shaping for truncation; conventional floating point DSP have significant and measurable noise floor modulation.
With EQ these problems are further compounded with IIR filters, as internal nodes are severely attenuated by low frequency coefficients - these nodes are then accumulatively amplified over time - thus a small error gets magnified into a substantial error. Indeed, I initially calculated that I would need 104 bits DSP for transparent operation - vastly larger than 64 bit DSP - but after designing the custom core, and running my suite of tests for transparency, the measurements for small signal phase failed - even with 104 bit processing on every internal node. Noise shaping allows the EQ to work linearly below 104 bits, as an error is constantly corrected. The core also passed my listening tests for transparency too.
Conventional 64 bit DSP suffers with measurable noise floor modulation, and significant errors for small signals - these filters fail to work with ultra-small signals.
I really like cross-feed - to me it's essential with headphone listening.
Note that the DSP is not parametric DSP, but a broad brush way of tuning the sound quality balance. That said, it can deal effectively with low frequency problems of headphones and IEMs.
I noticed a number of posters mentioning that Mojo was too loud with ultra sensitive IEMs. So the big change here is a bigger range for low volume. Also, pressing + and - together gives you a panic mute.
Pretty much the same as Mojo.
Again, no measurable noise floor modulation.
The only reason I could put in more features and better sound quality was by having better battery performance, and power efficiency - so I could spend that power budget on the FPGA.
The battery charging rate indicator is a great function. Most poor charging time is down to the cable, so being able to see the USB voltage is a useful feature.
When in desktop mode the battery is disconnected in order to maximise battery life. To do this I needed to improve the power supply rejection, by using my very low impedance discrete charger (when fully charged) and improved regulation for the amp. We can see this on the measurements as crosstalk (-118dB @ 1kHz) is better.
This project started in 2018, with many prototypes, to fine tune the performance.
It's going to be interesting to see how well the tone controls are accepted. As an audiophile I wouldn't touch EQ with a barge pole, as all implementations (analogue or digital) seriously degrades performance. So having a way of tweaking the sound to taste, or compensating for poor LF transducer performance, without degrading transparency at all, should prove to be a powerful tool. After all, how many of us have rejected something because it is simply too bright or too warm? Of course, if it's too bright because of distortion or noise floor modulation, then EQ can't cure that. But for a linear frequency response adjustment, EQ can help.
