Monday 7 October 2019

Paradigm Signature S8 v.3 Tweeter Replacement. A word about importance of speakers (+ rooms). (And a link to a HDMI cable blind test.)

Paradigm Reference Signature S8 v.3 tweeters.
I mentioned in late August with my "Speaker Impedance - Part 2" post that I noticed a variance between the right and left speaker impedance from my Paradigm Signature S8 v.3 that I was going to track down.

While an ohm here and there might look innocuous on the impedance curves, sometimes this could be indicative of a significant imbalance between speakers. Since I'm a bit obsessive about these things, let's get down to the "nuts and bolts" of the speaker and have a closer listen and look at the frequency response in that frequency region...

I turned off my usual room correction DSP, paid more attention to the the treble frequencies where the impedance looked "off" and indeed on a number of recordings, there were imaging issues with some vocals and instruments shifted "more left".

So what I did was using a laser distance measure for precision, positioned my Dayton Audio EMM-6 (with calibration) measurement mic 3-feet at tweeter level in front of the two speakers and measured the frequency response, overlaid with the graph of the impedance (frequency response is 1/12-octave smoothed, impedance curve 1/48-octave smoothing). This is all done with Room EQ Wizard software (v5.20 beta 24 & 25) and I used the Focusrite Forte audio device for ADC/DAC duties. The same Emotiva XPA-1L monoblock amplifier was used for each speaker with the same 12G speaker cables for each side. My listening level typically isn't very loud so let's target the volume to around 70-75dB SPL for these tests:

There's a bug in Room EQ Wizard 5.20 Beta 24 & 25- at the bottom legend, that should be 76.0 dB, not ohm. I forgot to change the color of the impedance to correlate with the FR but I think you get the idea.
Hmmm, interesting... And unfortunate. There's an area between 2kHz to 5kHz highlighted in yellow where there is a significant level difference between the two speakers, with the left louder than right. This should not be happening. At worst, such as around 2.3kHz where the cursor is placed, we see 5dB difference between the two sides.

Since the anomaly was affecting mainly frequencies in the upper midrange rather than a wholesale channel imbalance (which could be fixed with balance control), audibility of this abnormality varied depending on the material with content in those frequencies. When a female vocal for example reached towards 2kHz, the voice will drift in the soundstage slightly towards the left (this is where soprano female vocals like Sarah Brightman or Mariah Carey's vocal gymnastics on songs like "Emotions" were helpful). Likewise, hi-hats and the "shimmer" in cymbal swells often sounded like they're inappropriately shifted left of the drum kit (like on Dave Matthews Band's Crash).

Obviously something needed to be done! So I took out the tweeters to look for damage, compared and checked the crossovers which were accessible from the speaker terminals on the back. The crossovers looked fine.

Some speaker surgery... Tweeter replacement...
Indeed, it looks like the main difference was between the tweeters themselves.

So I contacted Paradigm's customer service department to buy a replacement beryllium tweeter (the right tweeter unit in the first picture up top is the new one, left is the older unit it replaced). I tried out the various combinations of left/right crossovers and tweeters (I ended up moving the original left tweeter over to the right and replaced the left tweeter with the new one), then did a repeat measurement similar to the graph above:

Better but not perfect. Remember that this is done with the microphone placed using a laser measurement device precisely between the two speakers with <1cm error. Our interaural distance is much larger than the pencil-width of the measurement microphone and our heads aren't affixed in a vise when listening so in reality we only need to be "in the ballpark" with these measurements! As you can see, there is now much smaller measurable speaker frequency response variation. I measured this 3 times and the one above is the "worst case" result I got.

Note that for the graph above, instead of the microphone being 3' in front of the speakers ("on axis" to the tweeter), the measurement above is with speakers measured from my usual listening position about 9' away sitting at the sofa and a few degrees off axis (speakers slightly toed-in). We can see that the replacement tweeter did affect both the impedance and improved the frequency response significantly.

At worst, the frequency response difference between right and left is now down to less than ~2.5dB (highest around 2kHz where the cursor placed) between 1.5kHz to 4kHz (highlighted in light yellow). The improvement is clearly measurable and most importantly audibly better compared with the older tweeter.

I typically listen with my subwoofer integrated in the system (Paradigm Reference SUB 1 which has internal DSP with room correction applied). This is what the right and left channel responses look like with the sub turned on, filling in the bass frequencies (again, measured from the listening position precisely between the two speakers):

The graph is plotted with 1/12-octave smoothing (1/6 and 1/3-octave settings will smooth over the valleys and peaks a bit too much IMO). Also, this is obviously in-room frequency response with various room modes at play and the effect of furniture in the room, not an empty "reflection free" and far from anechoic space.

The human ear is particularly sensitive to frequencies from 1.5-6kHz as per the Fletcher Munson Curve. This is why I was quite concerned about the frequency abnormality found and why it was worth getting this fixed (even if I could balance this out using DSP).

This is also a good time to remind everyone about the relative importance of speakers and room when it comes to high fidelity sound. Compared to the "cleanliness", "accuracy", and "ruler flat" response of a good DAC or amplifier, distortions introduced by speakers are high. For example, here's the calculated harmonic distortion graph for the right speaker + subwoofer (using Focusrite Forte DAC --> Emotiva XSP-1 preamp --> Emotive XPA-1L monoblock --> Paradigm Signature S8v3 / SUB 1), you can also see the noise level at 1kHz in the legend below:

As one often sees, the second harmonic is the strongest (red) and higher order harmonics are at lower levels although there's an interesting increase in 3rd order harmonic at around 1.5kHz.

Of all the pieces in a high-fi system, it is the speakers and room that are the most important in contributing to the final sound of the system. This is demonstrated in the distortion graph above. At 1kHz where the cursor is, we see that the 2nd order harmonic level is at 0.178% and THD (including 2nd to 9th harmonic) is at 0.292%. Even though the Dayton Audio small capsule electret microphone doesn't have the best SNR with resolution limitations, I suspect that we are "within the ballpark" of accurate measurements of total distortion for my system from Focusrite Forte DAC to speaker playback in my soundroom. Clearly this amount of distortion is easily orders of magnitude worse than a good DAC or amplifier measured in isolation; the vast majority of the distortion coming from the speakers and room interactions. As I expressed in my article over the summer "How Much Difference Does It Make?", remember to focus on speakers and rooms than DACs, streamers, even amplifiers (and generally let's not even talk about cables as being meaningful here!).

We can also do a realtime FFT analysis of the harmonic distortions. Here's 1kHz through the left Paradigm speaker using REW's signal generator:

I've found that the noise level measured in these realtime REW FFTs tends to be quite a bit higher compared to analysis of a recorded sine sweep.

Looking back, with regards to the abnormal tweeter that was replaced, I assume that the tweeter may have been damaged from use at some point between 2013 to 2015. Reviewing my blog archives, it looks like this difference between tweeters was present back in late 2015 when I did some measurements for room correction. Of course with room correction active, anomalies like this could be easily corrected for and hence I didn't notice the issue these last few years.

With the tweeter replacement, it also means I'll need to re-do my measurements for DSP room correction (perhaps another future post on this). As we can see from the graphs above, frequencies below 200Hz are where we should focus most of our room correction (the "Schroeder frequency"). Now that the tweeter anomaly has been addressed, I can leave the higher frequencies alone if I wanted to.

BTW, for those looking for more speaker measurements, make sure to also have a look at the SoundStage! Network's archive of measurements using Canada's National Research Council (NRC) anechoic facility. Lots of speakers have been measured through the facility over the years including graphs of impedance & electrical phase angle, on & off-axis frequency responses, and even THD+N curves. Of note, the Paradigm Signature S8 they measured back in 2004 seemed to be very different from the "v3" I have. I thought all S8's were ported designs yet the measured impedance curve looked like a closed box speaker. Strange.

Okay... Off to listen to some music now.


A friend recently got me hooked on Boris Blank's Electrified (2014); well done synth instrumentals à la Yello and much of it reminiscent of Convergence without the vocals. Check it out if you like synth-pop type stuff.

Finally, it has been good fun chatting with folks on E-mail in the last while. My apologies for not getting back quicker sometimes...

Remember a few weeks ago there was a discussion about digital cables and specifically HDMI. Recently I got an E-mail regarding a "blind" test of HDMI from a French magazine about a decade back (thanks Claude): Blind tests avec deux jurys : experts et lecteurs.

Turn on Google translate to read if needed. Basically, as expected no difference found between inexpensive 10€ cables and >100€ Monster and 75€ Belkin Pure AV cables back in those days among pros and forum readers. Cables were rated on both audio and video quality with average score out of 5 coming out essentially the same for each cable: 4/5 Belkin, 3.9/5 generic, 3.8/5 Monster.

They even wrapped the generic "no name" HDMI cable around a microwave oven and ran it while playing a movie! I guess the generic cable's shielding was just fine since they didn't see/hear a difference :-).

Until next time... Hope you're all enjoying the music.


  1. Thanks for publishing the graphs from your listening room, Archimago! Very interesting. Is it your preference to use 1/12 octave smoothing vs. "Psychoacoustic" in REW?

    For measurement microphone alignment I usually use Acourate's "Microphone Alignment" test. I think it's more convenient than with a laser meter, and maybe even more precise, because you can get down to 0 samples difference.

    1. Hi Mikhail,
      Yeah, I've just been using 1/12-octave, log scale as a standard that's easily accessible with most software rather than the psychoacoustic in REW.

      Thanks for the reminder here about Acourate's "Mic Align" feature. Indeed I've been using Acourate to remeasure my room and will post on this soon...

  2. As an enthusiastic DIY speaker builder, I always like 'speaker talking', so thanks for this very interesting article. I was surprised how much this small impedance difference could change the response. Like you, I measured frequency response differences between my left and right tweeter in this critical audio spectrum region, but as a DIYer, I made myself a tremendous gift when I bought a DEQX HDP-4 processor: with this extraordinary tool, I can calibrate the frequency response and time alignment parameters of each driver and therefore, that's the way I solved my the tweeter response differences. Question: since you wanted only to compare precisely the tweeters, any reasons why didn't you measure near field (and get rid of room reflections) ?

    1. Hi DColby,
      No good reason why not just do near field :-). Lazy I suppose since measuring the room response at the listening position meant I didn't have to move the microphone in front of each speaker; plus I just wanted to make sure where I was sitting was OK.

      I did measure at 3' in front like with the old tweeter as well and the difference was similar (did not show).

  3. Hi Archimago
    I came across Boris Blank at a hifi exhibition in the UK 2 years back and was smitten with it, highly recommended to those who like well produced recordings as much as electronic music.

    I would also recommend that you listen to Yello's album Toy, of which quite a few tracks were demos on electrified including the title track. Frautonium is a great opener, but overall the whole album is an exercise in perfectly produced electronic pop music.

    As you have great system I strongly recommend that you sit down for a full listen to the entire album, I can promise you that you will want to do some measurements of the DR and let others know, what an amazing recording it is.
    Only the recent Kraftwerk 3D catalogue release is as good as that, but I am biased towards electronic and well produced music.
    The most important thing is that all 3 examples I mention do take format and resolution out of the discussion and let the music really talk for itself. Whatever these guys did in the production process worked, regular go to listens.
    Recommended quick playlist
    Electrified Boris version
    Electrified Yellow version
    Blue Pacific
    Computer love

    Happy listening

    1. Nice Gordon!

      Always great to hear about music recommendations. I agree, Toy is a fantastically produced album and I've used some of the tracks over the years as well. Indeed, well worth listening start to finish.

      Hmmm, will have to check out the Kraftwerk!

  4. There's a known problem with the Soundstage measurements below 80 Hz or so. It's been remarked on at ASR.

    1. Thanks for the link Dave,

      Wasn't aware of this and a good read in that Hi-Fi+ article about the size of the NRC chamber. I'd imagine there's only so much one can do with creating some kind of compensation for the size limitation.

  5. Do the tweeters use ferrofluid?
    Also, once issue was isolated to tweeter , it might be advisable to do more of a nearfield measurement of just tweeter, perhaps from 1'. Do multiple sweeps and average, there will always be slight variations with SPL measurements.


    Soundfield Audio

    1. Hi AJ,
      Yes, it does have ferrofluid in the tweeters. I suppose I can try getting some and doing a "re-lube" of the one that's underperforming if that might help... Any tips?

    2. Unfortunately no, you did the right thing by replacing. I suspected it was ferrofluid issue. Keep an eye on the other one periodically.

    3. I'll certainly examine the tweeter when I have time, AJ to see whether I can replace the fluid and see if it normalizes...

  6. "I've found that the noise level measured in these realtime REW FFTs tends to be quite a bit higher compared to analysis of a recorded sine sweep."

    The N and THD+N figures from the RTA window are summed across the band, for that plot they will be dominated by the low frequency noise present. There isn't an equivalent for sweep measurements as it isn't possible to separate noise and distortion in a sweep and the graph is only showing results versus frequency. To provide some context for the sweep's distortion measurements there is a noise floor trace which is from a capture of the mic signal before the sweep starts. Distortion components which are around or below the noise floor are more likely noise than distortion. The RTA can measure much lower distortion levels than even a very long sweep, there is a stepped sine mode for accurate distortion measurements of electronic equipment and the like.

    1. Hi John!
      Greetings and thanks for the tip! I see what you mean about the RTA window. I'm only starting to appreciate the power of REW and looking to try out the stepped sine mode as well in the near future...

    2. Arch, as usual, love the article. Bummer about the tweeter. Good news on the outcome! I am sure AJ can sort you out on the ferrofluid... Just wanted to plug John's REW software. To me it is the defacto standard. With one acoustic measurement, there are so many ways to look at the data, it is a one stop shop. I love the ability to share .mdat's and compare using the overlay feature. I hope you explore REW further. Keep up the great work!
      Kind regards,

    3. Thanks for the note Mitch,
      Yeah, will have to explore REW a bit more... Have some ideas already :-).

  7. "They even wrapped the generic "no name" HDMI cable around a microwave oven and ran it while playing a movie! I guess the generic cable's shielding was just fine since they didn't see/hear a difference :-)."

    No surprise to me 'cause a microwave oven has a lambda shield.

    1. Hey unknown,

      Good point. But hey, even with microwave emissions blocked, "perfectionist" audiophiles and cable manufacturers would surely still obsess over the power supply noise, maybe physical vibrations from the oven itself causing "noise" in the digital signal, jitter, and stuff like that!

  8. Some people yes, they just don't get it.
    Btw, my name is Arnold and i'm from the Netherlands.

  9. Re: Paradigm tweeters.
    So much for tolerances & QA eh?
    I know Harbeth measures & matches speaker pairs & documents the serial numbers so that if you need replacement drivers you supposedly get an exact match for what you had.
    Perhaps at the price point of these speakers it isn't cost effective but I think these originally retailed for about $10k a pair.
    For that price you'd think the tolerances should be pretty tight.

  10. My story is a huge disappointment for you paradigm lovers. I have a set of S8 v2 and I think the people at paradigm haven't a clue how to build a tweeter. I have sent my be-tweets back 3 times and they keep coming back with a 2db or greater difference. You would think they can build a speaker or have some relatively decent quality control. NOT SO!!! bunch of idiots...

    I sought other means and now have a matched pair coming my way but WOW, what a mess. It only took 2 plus years....

    1. Hi AudioJunkie,
      Wondering how you managed to get the pair eventually? Was it still officially through Paradigm?

  11. is it possible to share the dealer contact details, I need to order replacement for my Paradigm S6 v.3 as well... facing the same issue here :(