imersiv D-1

Imersiv’s claims to have delivered a revolution in D‑A conversion technology aren’t mere snake oil: the D‑1 outperforms every other current DAC by a huge margin. And, importantly, the differences are audible...

Imersiv is a new name, but it’s a spin‑off from a very familiar company: Millennia Music & Media Systems. Founded by John La Grou in 1990, Millennia have been impressing SOS and many others with their high‑end gear ever since. John decided to create the new imersiv marque because their products will be based on a radically different technological concept from Millennia’s. In fact, as you’ll see below, it’s a radical departure from everyone’s products!

Their first release is the D‑1, a stereo digital‑to‑analogue converter (DAC). That might sound ‘ordinary’ to you, but I assure you this converter isn’t! It outperforms every converter that I’ve tested previously, and by a considerable margin. It’s not just about tech‑spec bragging rights either: there are clear, audible benefits to imersiv’s innovative approach. Yes, it’s eye‑wateringly expensive, but don’t let that put you off just yet: this is the sort of technological breakthrough worth being aware of, and which should in time filter through to products that we’ll all use.

Technology

Before I dive into the details, it’s probably helpful to put imersiv’s achievement into context. Back in 1887, Edison’s original cylinder recorder had a usable dynamic range of just 15dB. The first electrical sound recording and replay systems (to 78rpm disc) raised the benchmark to about 40dB, and magnetic tape, widely adopted in the early 1950s, was another big step up in quality, increasing it to around 65dB. In the early 1980s, the industry adopted 16‑bit digital audio, which raised the potential dynamic range to about 90dB (soon to be largely wasted in the ‘loudness wars’!). Each of those major technological leaps increased the dynamic range capability of both recording and playback systems by roughly 25dB.

But while we’ve recently seen 32‑bit floating‑point converters appear on the recording side (more on that parallel later), when it comes to D‑A conversion there’s been only a continual stream of small, incremental improvements. In fact, averaged over the last 45 years or so, the dynamic range of D‑A converters has improved by just 0.8dB per year — that’s roughly the same as the continuous average rate of dynamic range increase since 1887!

Don’t get me wrong. Those improvements result in a modern crop of high‑end D‑A converters that is good. Very, very good, in fact! The best I’d measured before now achieves a dynamic range of around 125dB, which is the equivalent of 21.5 bits of ‘resolution’. Looking across the league table of D‑A converters that I’ve personally tested, the top seven all achieve AES17 dynamic range figures of over 124.5dB, and they all sound superb. Nonetheless, the development of DACs with more than 22 bits of real dynamic range has proven a stubborn technological challenge.

The real practical benefit is a natural side‑effect of that vast dynamic range: a substantial improvement in distortion levels for mid‑ and low‑level signals.

It’s a challenge to which imersiv have risen — it’s also a ceiling they’ve shattered! They claim that the D‑1 DAC’s ‘HDR‑A’ technology delivers 28 bits of real dynamic range, which equates to an extraordinary 168dB. That’s an astonishing increase in dynamic range of around 40dB, and genuinely a step change. But that headline‑grabbing dynamic range figure isn’t even the main benefit! After all, who can realistically listen to anything with more than about 130dB dynamic range without causing damage? The real, practical benefit for those listening is a natural side‑effect of that vast dynamic range: a substantial improvement in distortion levels for mid‑ and low‑level signals, over even the very best conventional D‑A converters.

This diagram, provided by imersiv, illustrates the digital gain‑mapping concept that lies at the heart of this technology. Effectively, the aim is to ensure that the signal is always in the optimum operating range of the converter chips, whatever the actual signal level.

In a conventional converter, as the digital signal level reduces, so too does the signal‑to‑noise ratio; the wanted signal becomes closer to the dithered noise floor. Equally important is that the total harmonic distortion (THD) rises as a proportion of the wanted audio signal (hence, manufacturers always specify converter THD at 0dBFS, where they obtain the smallest possible figure!). This matters, because a lot of recorded music spends much of the time in those mid‑ and low‑level regions — and substantially lower distortion there is experienced as the proverbial “removing of a veil between the source and the listener”. Put simply, it brings more vivid clarity and naturalness to the recorded sound.

It’s fair to say, then, that the D‑1 is a revolutionary, rather than an evolutionary, technology, and that imersiv are fully justified in calling it a ‘paradigm shift’. So let’s now look more closely at how they’ve achieved this...

The Photography Analogy

Imersiv’s new technology is called HDR‑A, which stands for High Dynamic Range Audio. The terminology is borrowed from high dynamic‑range photography and, while not quite a perfect analogy, there are useful similarities. HDR photography typically blends high‑, medium‑ and low‑exposure images in a clever way, to capture the enormous dynamic range of real life when using camera sensors that have insufficient dynamic range capability to cope unaided. In this way, details are preserved from both the darkest and brightest areas, as well as in the mid tones. Now, photographic HDR is, technically, a form of dynamic range compression — so before anyone starts panicking, I want to stress that there’s no dynamic audio compression involved here. HDR‑A is simply borrowing the idea of using multiple sensors, each optimised for different parts of the dynamic range.

In essence, HDR‑A employs multiple independent D‑A signal paths, each for a different sector of the source’s dynamic range. In theory, any number of paths could be combined, but to keep the explanation clear let’s consider just two: a high‑level path and a low‑level path. The analogue output is derived by combining the signals from the separate D‑A converters used in each path, to deliver an overall dynamic range that’s far greater than any conventional single D‑A stage could ever achieve. More importantly, this configuration also achieves much lower (like 40dB lower!) noise and THD than ever before.

Now, as I explained, the THD always rises as the digital input level falls. The idea behind the multiple paths tactic is that, before the THD rises too far in the high‑level path’s converter, the HDR‑A technology switches the digital input across to the second (low‑level) converter path. To optimise the performance of this second converter, this lower‑level portion of the digital input is bit‑shifted upwards, so that it feeds the second converter at the top of its own dynamic range window — there, the THD is again extremely low. Consequently, the THD across the mid and low range of this HDR‑A converter is much lower overall.

This concept might remind you of the multiple A‑D converter configurations I mentioned above, which have become de rigeur in portable floating‑point digital recorders in recent years. Yet, while the concept is similar, it is also different in significant ways, and there’s one particularly important distinction. Although multi‑path A‑D converters currently deliver 32‑bit floating‑point outputs, a floating‑point signal has only 24 bits of absolute signal resolution; the rest of the floating‑point data is used to scale those 24 bits up or down within a vast dynamic window, to provide almost unlimited headroom and an incredibly low processing noise floor. That’s not technically the same thing as a contiguously expanded dynamic range. In contrast, the D‑1 can accept fixed‑point audio with up to a 32‑bit word length — and in that case, it provides a contiguous 190dB of dynamic range (assuming the presence of dither noise).

Devil In The Details

While the broad concept behind the HDR‑A technology might seem fairly straightforward (indeed, its origins can be traced as far back as 1981), the devil is, as always, in the most intricate of engineering details. Hence, it took Millennia a full decade of R&D, huge investment, and the acquisition of three patents before they reached the stage where a commercial product was viable. So, let’s look at some of those intricate details...

The example I’ll use is derived from La Grou’s 2021 AES paper (https://aes2.org/publications/elibrary-page/?id=21106#), in which he describes an early prototype of the D‑A converter. At the time, it was achieving a total dynamic range of 156dB (26 bits) and a ‑140dBu noise floor. The current D‑1 improves on that, with 168dB of dynamic range (28 bits) and a ‑146dB noise floor — this is due to the more advanced converters used in the two signal paths, and tweaks to the signal‑path topology. To help you understand what’s going on, I’ll refer to the MSB (most significant bit: ‘the loud end’, if you like!) as bit 1, and the LSB (least significant bit/quiet end) as bit 32. Some might consider this to be ‘backwards’, but it makes it easier to relate to the word lengths being fed into the 24‑bit converter chips in each signal path.

First, a DSP processes the 32‑bit fixed‑point digital input signal, to create two separate digital streams. In that initial digital processing, the bottom six bits, (32‑27) are effectively discarded because they will be well below the noise floor of the low‑level converter (and of all analogue circuitry). The top seven bits (1 to 7) cover the highest 42dB of the signal’s dynamic range, and are routed to the high‑level converter, so use its region of best THD performance.

The analogue output from the high‑level converter is passed through an amplifier to boost it by 8dB, and raise the peak output level to +16dBu (as per the AES paper; in the D‑1, the peak output level is a more professionally acceptable +22dBu). The remaining 19 bits of the digital input (bits 8‑26) are all shifted upwards by seven bits (ie. 42dB of digital gain) before being dispatched to the low‑level converter’s path, once again optimising the THD performance through that converter.

Obviously, in theory, the specific break point where the digital signal is split between the two converter paths could be set at any desired level, to fully optimise the performance of whatever converter chips are used. So too can the amount of gain added to the high‑level path; it’s all part of optimising the signal‑to‑noise and THD performance of the complete system.

Also worth noting is that, although I’ve suggested the digital input is hard‑switched at a set level, the signal is actually crossfaded between the two paths over a small range, to avoid undesirable level jumps, and a self‑calibration process allows the DSP to monitor the two signal paths and adjust the crossfade algorithm to minimise any level offset.

There’s yet more complexity there, too, because the DSP must know in advance how the signal level will vary, if it’s to ensure that the crossfade avoids overloading the low‑level path, and if it’s to prevent the signal remaining on the high‑level converter path too long, which would degrade the performance. So the DSP also delays the input signal, to allow a look‑ahead facility. The AES paper states that a 1ms look‑ahead delay is sufficient but, while the D‑1 can match that if required, its default setting is 20ms.

Mixing Highs & Lows

The engineering challenges don’t end there, though, and the next one is how best to combine the outputs from the high and low converter paths, to form the single analogue output signal. Since the digital signal was shifted upwards to feed the low‑level path, the analogue output from that converter must be attenuated by the same amount (42dB in the example), to restore the correct amplitude relative to the high‑level path.

Part of that level shift was effectively achieved by the 8dB boost of the high‑level path, leaving ‑34dB to be applied using a passive (resistance) attenuator. That’s not as trivial as it seems, because the attenuation must be incredibly precise (to within one LSB), plus stable with temperature, if the linearity of the whole converter is to be maintained.

In attenuating the low‑level path’s analogue output, the converter’s residual noise floor is also attenuated, of course, and that’s why the system noise floor is so incredibly low. The passive attenuator does inherently generate some thermal noise of its own but if its resistance values are kept relatively low, this is insignificant in practice. The AES paper quotes a residual noise floor in the prototype dual‑path converter of just 77 nano‑Volts, which translates to ‑140dBu. That’s already a spectacular result that’s well below the noise floor of most conventional line‑level analogue electronics. But as discussed above, the D‑1 exceeds even this, with a noise floor that’s claimed to be 6dB lower at 40nV (‑146dBu).

A small snag is that while the low‑level path’s noise floor is incredibly low, thanks to 34dB of passive attenuation, the high‑level path’s converter output isn’t attenuated. On the contrary, it’s being amplified by 8dB to achieve the required peak output level. So, if the high‑ and low‑path outputs were simply mixed, the noise floor of the high‑level converter would swamp that of the low‑level one, destroying the whole aim of this dual‑path topology. The only way to overcome that is to switch off the high‑level path’s contribution whenever the digital signal is routed to the low‑level path. Imersiv achieve this through a muting gate on the high path’s analogue output, controlled by the DSP. In the AES paper prototype, the high‑level path is muted when the audio signal falls below ‑42dBFS (when it is routed to the low‑level path), to maintain the amazingly low residual noise floor for mid‑ and low‑level signals.

Of course, when the input signal is higher than ‑42dBFS (so the high‑level path is used), the system noise floor is determined by the high‑level converter. So it jumps from ‑146dBu to about ‑104dBu, depending on the amplitude of the input signal. That’s obviously not ideal, but thankfully there’s a ‘Get Out Of Jail Free’ card: psychoacoustics! Viewed on its own, the high path is a conventional 24‑bit converter, calibrated to provide a professional line‑level output, with a noise floor around ‑104dBu. That’s much the same as the best current converters, and this noise floor is always going to be at least 80dB below the quietest audio signal passing through that converter. The human auditory system suffers from what, in this case, is a very handy characteristic called ‘noise masking’: in the presence of a much louder signal, we simply cannot perceive broadband noise, which means the jumping system noise floor level isn’t at all noticeable!

Looking at it another way, the absolute worst‑case signal‑to‑noise ratio with this technology occurs for transitional mid‑level signals, where the noise floor is the same as in the best conventional single‑path DAC. But quieter signals, below the transition level, benefit from substantially better signal‑to‑noise and THD performance.

OK, I’m aware that I’ve offered a lot of numbers above! The practical upshot of this innovative technology is this... If you were to generate a digital sine wave at ‑100dBFS and feed it into a conventional D‑A converter, the analogue output would look pretty noisy and distorted. Feed that same signal into the D‑1 and it comes out looking impressively undistorted and pretty much noise‑free. And it translates to a cleaner, more accurate, more faithful reproduction, particularly of mid‑ and low‑level sounds, that is most definitely audible.

In The Flesh

Happily, while the underlying technology may be complicated, the D‑1 isn’t so from a user point of view: in fact, it’s very straightforward to connect and operate. It’s housed in a compact standalone chassis, with a white case and your choice of silver or black front panel. It measures 218 x 81 x 414mm (WHD (for my American friends and Brits of a certain vintage, that’s 8.6 x 3.2 x 16.3 inches). It weighs a modest 3.4kg (7.5lbs), and optional rack ears are available for mounting one or two D‑1s in a 2U 19‑inch rack space. Power is provided from an external line‑lump SMPS unit that accepts 90‑240 Volts AC mains, at 50‑60 Hz, outputs 12V DC for the D‑1’s circuitry, and connects using a mini‑DIN plug. A miniature toggle switch on the rear powers the unit on or off, and it draws a nominal 15W of power.

The D‑1 caters for 32 to 384 kHz digital input signals with any fixed‑point word length up to 32 bits, over USB and S/PDIF, both coaxial (BNC) and optical (Toslink). DSD playback is also supported, but the signal is converted internally.

On the right‑hand side of the rear panel are pairs of balanced analogue audio outputs (male XLRs) and unbalanced analogue outputs (RCA phono), but at the time of writing the latter are described as being for ‘utility only’ and their technical performance remains unspecified. Moreover, the RCA phonos can’t be used when the XLRs are in use.

The manual states that the published specifications are only valid for the balanced outputs, which must be connected to a true differential destination (a transformer or electronically balanced input) to ensure the D‑1’s output topology performs correctly. Because the summing of the high and low paths is affected by the impedance of the cabling and destination, once you’ve connected the outputs to a suitable destination, you must run a calibration sequence (more on that below). The maximum output level from the balanced XLRs is +22dBu, from an output impedance of just 1.8Ω.

The digital audio inputs, on the left‑hand side, sit behind a thick, protruding metal fin that extends through the unit to shield the digital electronics from the analogue circuitry (an important consideration, given the extremely low analogue noise floor). The digital inputs comprise three 75Ω BNCs for coaxial S/PDIF and a Toslink port for optical S/PDIF. (Most S/PDIF sources use RCA phono connectors, so a trio of suitable adaptors is included.) There’s also a female XLR for AES3, and a USB‑B port for connection to a computer, for class‑compliant audio replay as well as firmware updates. Dante networking is available too, as a factory‑fitted cost option, with both primary and secondary RJ45 sockets. All standard digital sample rates from 32 to 384 kHz are supported, with any fixed‑point word length up to 32 bits. DSD64 to DSD256 are also supported in the DoP format, and all DSD sources are converted to PCM at 192kHz to facilitate the multi‑path DSP processing.

The elegant front panel hosts a standard quarter‑inch headphone socket, a large rotary encoder, a central 256x64 pixel OLED screen in the centre, and five illuminated buttons. These choose the input source, access the setup menu, select a menu value, access the output filter menu and engage an output mute. A fully balanced headphone output (on a Pentaconn connector rather than the usual TRS socket) is available as another factory‑fitted cost‑option.

Operation & Calibration

When not used to access menu options, the encoder wheel also adjusts the output volume in 1dB increments (or 0.1dB steps by holding the Select button while rotating the knob) over a 154dB range (scaled ‑130 to +24). The volume control is performed on the digital input signal by the DSP, and 0dB on the volume scale actually equates to +21.7dBu for a 0dBFS input. In other words, there is up to 24dB of gain available for monitoring low‑level recordings, but any positive volume setting over 0dB risks overloading the high‑level converter if the input hits 0dBFS.

The default display shows the selected input source and sample rate in the bottom centre of the two‑line display. The current output volume is given above, and around the sides various code letters indicate other active functions: H for Headphone Exclusive, M for Mute, P for inverted Polarity, R for Resampling mode, and Fn for the various reconstruction Filter options.

Pressing the Input button opens the input source menu, and the encoder wheel selects the desired option. The Filter menu works the same way with the standard filter mode, F0, applying a linear‑phase reconstruction filter with a fast roll‑off. Other options comprise linear‑phase with a slow roll‑off (F1), linear‑phase apodising (F2), and three minimum‑phase variants: fast (F3), slow (F4) and hybrid roll‑off (F5).

The Setup menu contains various sub‑menus, the first option being the Calibrate process I mentioned in passing above. This is used once the D‑1 is connected to a suitable differential (balanced) destination, minimising the amplitude error that could potentially occur as the signal is crossfaded between the high‑ and low‑level signal paths, because of the susceptibility of the passive summing to output load impedance variations. Separate calibrations are performed (and stored) for the headphone and balanced line outputs.

During a calibration cycle, which lasts up to 30 seconds, the D‑1 emits various digitally generated ‘pilot tones’ through both internal paths, measuring the voltage offset between them. Based on the measured error, the DSP adjusts the crossfade algorithm to reduce the error, and repeats the process to make ever finer path‑matching adjustments. The process completes when the error is below 0.005dB: the display then returns ‘Successful’ or ‘Failed’ messages. In the latter case, it will revert to the last successful DSP values. Obviously, if the D‑1’s outputs are connected to a different destination or via different cables, the calibration process should be repeated.

A Polarity menu option simply inverts the output signal polarity, while Power Off puts the unit into standby mode, from which it’s awoken by pressing any button or turning the encoder. The Firmware menu displays the installed firmware, while the Main Menu page allows the display to be customised between Normal, Logo, or None options. Normal is as described above, while None turns the display off completely. The Logo option simply displays imersiv’s logo.

Low Latency is a yes/no choice, with the default No incurring a 20ms latency. In applications where half a video frame of audio delay might cause problems, Yes reduces latency to under 1ms, but there’s no audible difference in quality. Resample is another yes/no menu: choosing Yes automatically converts 32 to 176.4 kHz digital inputs up to 192kHz for all further processing and conversion; select No, and processing and conversion are performed at the native sample rate. The yes/no Headphone Exclusive menu item does exactly what you’d expect.

There’s also a hidden ‘Pro Menu’ page, accessed by pressing and holding Setup and Select. It contains options to adjust the left‑right Balance (±6dB in 0.01dB increments) and the Mid balance (±12dB in 0.1dB increments) — this being the Mid component of a Mid‑Sides conversion so, essentially, it’s a stereo width control. There’s also a Level Lock mode which disables the adjustable output volume.

You might think the final option unusual, given that imersiv have invented a whole new way of minimising harmonic distortion in a D‑A converter, and I’d agree: it intentionally adds different blends of (mostly even) harmonic distortion, to add ‘richness and warmth’. Apparently this function, which can be used to personalise the sound character with (subtle) musically enhancing coloration or completely disabled as you prefer, was added during the beta testing at the behest of a mastering engineer who enjoyed the musically enhancing character of his previous DAC.

The included infra‑red remote controller’s large +/‑ buttons adjust the volume, and all seven input sources can be selected using their own buttons, as can all six filter modes. Additional buttons select mute, polarity‑normal/polarity‑invert, headphone and line out, and display on/off. Four buttons labelled R1‑4 are apparently reserved for future use.

Listening

Finally, I should let you know how the D‑1 actually sounds! Well, having hooked up the D‑1 in my studio, I was initially somewhat underwhelmed (and I stress: initially!). It sounded neither quieter than my Crookwood or Benchmark converters, nor any more dynamic at my listening levels. So I took it into my living room and connected it into my Bryston/PMC hi‑fi, replacing the DAC in my Bryston BP173 preamp and... still no real difference.

But as I drank my Masala Chai and worked through my standard test tracks — material that I know in intimate detail — I started noticing things that I simply hadn’t before: low‑level details became more precise, more real, and less blurred or veiled. The more I listened, the more obvious this became. Well‑recorded tracks gained a vivid realism and, often, individual instruments sounded like they were genuinely in the room rather than on a recording. The difference was mainly in the starting transients and decays, particularly in stringed instruments, both bowed and plucked. It was also there in the attack and decay of cymbals and hand drums — and, especially, in the realism of vocals.

When listening to good recordings made with simple stereo mic arrays, the room acoustics demonstrate significantly more realistic depth and width, too: each instrument occupies a far more clearly defined position within that space. I also heard so many tiny details on some of my own recordings that previously I hadn’t even realised existed. It’s a cliché, I know, but it really was like someone had cleaned the audio window and I was making out some fine details for the first time!

Moving the unit back into the studio (and recalibrating it), I could now hear that same vivid realism and precision. It transpired that the lack of noticeably reduced noise floor had been because the D‑1 was routed through the Crookwood mastering console’s monitoring path — and the console entirely defined the system’s noise floor!

What Price Progress?

Groundbreaking products never come cheap, and the imersiv D‑1 is no exception: it costs an eye‑watering £12,000$12,000 USD, and the Dante and balanced headphone output options each add around $1000. Clearly, this is not a D‑A converter for the average project studio or back‑bedroom beats creator, but it will be within the reach of high‑end studios and mastering rooms. The important point, though, isn’t the price. Rather, it’s that the D‑1 proves that this concept actually works, and that there are practical, real‑world benefits.

It is a technical tour de force that raises the audio benchmark dramatically.

It is a technical tour de force that raises the audio benchmark dramatically. I have no doubt — with imersiv exploiting their IP, other manufacturing approaches (integrating everything into a single chip?) becoming viable, and other manufacturers, in time, developing their own equivalent solutions — that, a decade from now, this kind of technology will be the standard way of converting digital audio to analogue, and we will all be able to enjoy the associated sound quality benefits. John La Grou and his boffins really have moved the goalposts, and they fully deserve all the professional accolades that I’m sure they’ll receive!