Waiting for a mate. Recent years have seen the KEF LS50 hold their ground at (close to) the top of goto lists for loudspeaker auditions in the sub-US$2K space. Rolled up inside an enclosure that makes a break from the aesthetic conservatism of the past sits KEF’s Uni-Q driver configuration: a 2.5cm aluminium dome tweeter coaxially aligned with a 13cm magnesium/alloy bass-mid driver.
Not having heard every other standmount in and around the LS50’s price point, I might not comfortably assert ‘best’. What I can say with confidence is that the baby KEFs are popular. Four years on from the LS50’s (initially limited) release, this loudspeaker continues to resonate with the buying public — maybe because, when we boil it down to bare essentials, they sound good and they look good.
The LS50’s talents with punch, poise, clarity and (especially) imaging have seen them as a DAR staple for the best part of three years. Last year I traded in an original “50th Anniversary Edition” black pair with the gold Uni-Q driver array to the limited edition Racing Red finish with the black/silver woofer/tweeter; a matter of appearances and nothing else.
This week though, the LS50 sound quite a bit better than they did last week. From Grace Jones to The Flaming Lips to Ancient Methods, the Red Racers present with a fuller midrange, a more refined top end and mode cleanly defined player outlines.
What upstream change could possibly net such a pronounced result?
The amplifier remains the same (a Devialet Expert 200). Ditto the D/A converter (again the Devialet). The loudspeaker haven’t changed – still AudioQuest’s Rocket 88 – and the connection between Roon Endpoint box (an Antipodes DX Gen 2) and the Devialet remains a Blue Jean Cable S/PDIF coaxial.
That leaves only one possibility: a change to the Roon Endpoint hardware. However, this switcheroo isn’t simply a matter of lowering noise or jitter but the digital domain execution of something more radical: loudspeaker correction.
As good as the LS50 is, it isn’t perfect. No loudspeaker is no matter what the marketing materials tell us.
“Speakers do their best to convert the voltage representing amplitude to a precise air pressure arriving at our ears at any given instant in time. Compared to the upstream electronics that drives them, loudspeakers must integrate several uncompensated mechanical transducers (bass, midrange and tweeter drivers) as transparently as possible and it’s very difficult to get that exactly right.”
“Every speaker driver introduces errors to varying degrees whereby the sound they produce differs from the signal driving them at certain frequencies or frequency groups. Some frequencies will play slightly louder or softer than specified (frequency response errors) and be variously delayed compared to others (group delay errors).”
Those are the website words of Australia’s DEQX (pronounced “Decks”) whose PreMATE+ digital processor (US$5995) can pre-emptively correct for a loudspeaker’s frequency and timing errors. In other words, the LS50’s 2-way driver output can be improved.
“If we correct for group delay (timing errors) then any frequency response anomalies are also corrected – the two are mathematically intertwined”, says DEQX General Manager Alan Langford.
The Sydneysider elaborates: “It [the DEQX PreMATE+] corrects speaker frequency-response and timing errors by adjusting thousands of frequency groups, depending on your speakers’ measurements, so that they arrive at the correct time. Traditional analogue and digital EQ corrupts the timing coherence around the frequencies that they’re trying to equalise for volume.”
“DEQX Processors slow down on-time and early-arriving frequencies so that slower frequencies can catch up. DEQX uses real computing horsepower to make that happen with a 240-megaflop, 32-bit floating-point Digital Signal Processing Engine that essentially achieves zero distortion.”
If that causes your brain to perform hitherto unwitnessed mental gymnastics then you’re not alone. As with many things in hifi, one doesn’t necessarily need to fully understand the engineering theory to enjoy the end result.
Think of it this way: just as eyes are metaphorical windows on the soul, our loudspeakers are windows on upstream electronics. The way Langford tells it, loudspeakers introduce more distortion than any other component in the system; frequency and timing (group delay) errors erode ultimate output resolution. Correct for those errors and some resolution can be reclaimed.
Frequency errors means some frequencies are reproduced more loudly than others. The loudspeaker engineer’s pursuit of a ruler flat frequency response – all frequencies reproduced at the same volume – is but a pipe dream. (Discussions on the validity of ‘flat’ as as the optimal target remain well beyond the scope of this coverage)
Group Delay on the other hand refers to groups of frequencies arriving at the ear before others. These are sometimes referred to as timing errors. We don’t want that either
DEQX’s proposition is to apply a long, linear phase FIR filter to correct for group delay anomalies. As we shall see, this is turn evens out a loudspeaker’s measured frequency response.
Compensating for group delay errors can, according to Langford, only be executed in the digital domain. DEQX’s PreMATE+ applies a digital filter to the signal before it reaches the D/A conversion stage; either the PreMATE+’s own internal DAC (which in turn spills its analogue output via balanced XLR or single-ended RCA socketry) or a third party decoder connected to the unit’s solitary digital output (S/PDIF BNC).
“We don’t publish which model DAC chip is used in the PreMATE+”, says Langford. Moving on…
On the PreMATE+’s digital inputs, we count five: TOSLINK, S/PDIF BNC, S/PDIF RCA, AES/EBU and USB. As alluded to above, the PreMATE+’s official stamp of Roon Readiness is imminent. MPD compatibility is also offered.
In (over-)simplified terms, we might describe the PreMATE+ as a pre-amplifier with digital domain volume attenuation. We might then incorrectly infer that vinyl spinners and tape spoolers aren’t invited to the party. But they are. A 24-bit Cirrus Logic ADC running at 96kHz greets analogue signals arriving at the DEQX box’s unbalanced RCA and balanced XLR sockets.
Analogue purists might bristle at the thought of turning over their phono stage’s or reel-to-reel’s output to ones and zeroes but ultimately such prejudice is their quite literal loss (of resolution). The audible gains from DEQX-correcting the KEF LS50 more than offsets any real or imagined ‘analogue magic’ surrender.
Back at DEQX’s online FAQ: “Normally, this high resolution impulse response convolution adds around a quarter second (250 milliseconds) to process, so synchronising a ‘corrected’ soundtrack to video for example is a problem to say the least. DEQX maintains high resolution adding only around 5% of the time usually taken, or about a quarter film frame’s delay of the audio that’s not noticeable.”
In dodging unnecessary latency, a pair of high power SHARC DSPs run the 500,000+ lines of DEQX-Cal™ code. These are not to be confused with the ARM A8 processor is charged with a) marshalling USB and Ethernet data and b) running the 480px x 270px touchscreen that offers an additional layer of hands-on device control over and above the more basic remote wand. Virtual VU meters too!
How is the loudspeaker correction linear FIR filter calculated?
For those not comfortable going it alone with the 172-page user manual, the company’s DEQXpert service brings online support via Teamviewer and Skype. You move the microphone, DEQX takes care of the rest. Stateside DEQX-ers enjoy time-zone appropriate support.
I had the home turf advantage of living 25kms from DEQX’s Sydney HQ. Alan Langford delivered the PreMATE+ review unit in person and guided me through the basics. Plug and play it is not but you get out what you put in – literally and figuratively.
Be advised that what follows extended my knowledge of loudspeaker design and functionality to new limits. To say that I learnt a lot is an understatement.
The basic process is this: measure the loudspeaker, have DEQX’s Calibration software calculate its correction filter, load it into the PreMATE+ over USB and then re-measure the loudspeaker in order to verify the filter’s corrective capabilities. Measure – Calibrate – Verify. 1 – 2 – 3.
The heavy lifting is carried out by DEQX’s own Windows-only software application which, connected via USB to the Pre-Mate+, interprets the reading of frequency sweeps made by a microphone connected to the DEQX box’s XLR socket. Langford supplied the Dayton Audio EMM-6 (US$79.99) and mic stand as loaner items.
For an anechoic-like measurement the microphone’s nose is placed 60cm or so in front of the left LS50, itself standing in as much free space as possible.
The Calibration runs several frequency sweeps before reporting back with the KEF standmount’s impulse response:
From this impulse response measurement the DEQX Calibration software derives frequency response:
Note that the DEQX-measured curve is entirely consistent with KEF’s own sheeted specification for the LS50: 79Hz – 28kHz (±3dB). In other words, between 79Hz and 28kHz the red line wanders not above +3dB or below -3dB. According to Langford, the peak at 40kHz is most likely the tweeter’s resonant frequency.
Group delay is also calculated:
Here we see timing errors really begin to swing with frequencies below 200Hz. Langford suggests that this probably isn’t the loudspeaker’s doing but more likely an indication that the room has begun to interfere with the near field measurement.
Looking back at the impulse response (above) confirms as much. Note the squiggle just beyond around 10ms? That’s the room coming into play. Throwing it back to high school science, we recall that time corresponds to frequency: 1000ms equates to 1Hz. Similarly 10ms converts to 100Hz.
Keeping it safe, Langford chooses to have the PreMATE+ gate the loudspeaker correction at 200Hz. Only frequencies above 200Hz will enjoy the Calibration software’s auto-calculated fix up.
[Side note: the bigger the room, the lower the frequency at which it comes into play and therefore the wider the bandwidth that can be corrected for group delay errors.]
Once calibrated the filter is loaded into one of four preset locations on the PreMATE+. Remote control access to these four positions means A/B-ing from the listening position is a snap.
You already know my thoughts on how much the corrected LS50 sounds – but how does it measure? Time to verify by engaging the correction filter and re-measuring.
Group delay above 200Hz comes out like this:
Moving from the original measurement in red to the corrected measurement in blue, we note a more evenly distributed group delay.
What implications for the LS50’s frequency response? Original in red, corrected in blue:
DEQX corrected, we get a smoother, more even response from the LS50.
These measured improvements seem to correlate with the subjective listening assessment that the KEFs now sounded better.
Verifying the correction filter’s SQ lift by subtracting it confirmed as much. Without DEQX FIR filter intervention the LS50 present as a few kilos lighter from the female voice region down to the midbass but also less settled/organised up top. Accusing the uncorrected LS50 of less high frequency extension would stand up in an audio court (should it exist) but it’s not quite as simple as that – it’s also less coherent.
The DEQX PreMATE+’s loudspeaker correction proves itself on screen and in-ear that it can take an already great loudspeaker like the KEF LS50 and make it sound (and measure) even better.
The audible improvements are more pronounced than switching up cables, server/streamer, DAC or amplifier. Such is the power of low-latency group delay correction through DSP. Australians say “Maaate”, I say “Bloody marvellous!”.
And yet, this DEQX box’s software-hardware one-two does quite a bit more than just loudspeaker correction.
Further information: DEQX