Dual Compression Drivers using Y-Adapter VS Celestion Axi2050 and RCF ND950

Dual Compression Drivers using Y-Adapter VS Celestion Axi2050 and RCF ND950


In this blog post I compare three different compression drivers. 

  1. Celestion Axi2050
    1. Mounted to Horn No.1723 
  1. RCF ND950 2.0
    1. Mounted to Horn No.1723 (same as above)

  1. Dual SB Audience 65CDN-T with Y-adapter and custom rear chambers 

Horn No.1300 is used on the Sabourin 1309 speaker plans. The Y-adapter bolts directly to the existing design. Those that have built the 1309 plans can easily change the configuration to the dual 65CND-T. This solution covers the same bandwidth as the GRS planar but with much higher sensitivity and output capability. I do not have a crossover developed for this configuration yet. Please contact me for inquires.  

The RCF and Celestion are tested on horn No.1723. I currently do not offer the 1723 horns plans on my site. If you are interested in the plans for this horn please contact me. 

Why this comparison?

Recently I've been testing large format compression drivers including the Axi2050 and RCF ND850. The Celestion did well on the 1723 horn providing low harmonic distortion through the midrange up to 2kHz. 

So I purchased the RCF ND950 (4" voice coil) version to see how it would fare against the Celestion in a direct comparison. 

I decided to add a wild card into the ring, a dual driver using a Y-adapter. These 2.50" voice coil compression drivers could potentially match the performance of the larger 4" voice coil drivers. The dual driver configuration has close to 10 Sq inches of diaphragm radiating surface area, while the 4" voice coil has 12.50 square inches...so pretty closely matched. As an added benefit, the 2.50" voice coil on the 65CDN-T has bandwidth up to 14kHz where the RCF has diaphragm breakup starting at 8kHz. So if the dual 65CDN-T configuration matches the performance of the other two drivers then it is a win win for wide bandwidth and ultra low distortion. In order to lower the driver's FS for low midrange duty I have replaced the existing rear cover on the 65CDN-T with custom 3D printed covers which increase the rear chamber volume. This lowers the FS from 480Hz down to 279Hz. I've summarized the potential advantages of the dual 65CDN-T below...

  • Wide bandwidth coverage from 300Hz-14kHz (breakup free)
  • Wide horizontal off-axis coverage due to the narrow vertical slot width of 1.40" (The Sabourin Horn polar characteristics can be found here)
  • Very high sensitivity of 120dB a 2.83V, 4 ohm nominal load.
  • Very high maximum output SPL 

Below is a picture of the dual 65CDN-T drivers mounted to my custom 3D printed Y-Adapter (reference part No. 2016-01-001). I also designed custom rear covers (reference part No.2016-01-002).

Shown below is the custom 3D printed Y-Adapter and rear covers. 

Celestion Axi2050

I begin by measuring the frequency response of the Celestion Axi2050 on Horn No.1723 which has a 200Hz cutoff frequency. I've had to add an 8ms gate to kill reflections in my small studio. The reason I'm measuring in my studio is so that I can get the room extremely quiet for the distortion tests. The Celestion exhibits a relatively smooth response although it has a -6dB/octave falling response starting at 2kHz. 

If I remove the gating we see a little more resolution near the cuttoff of the horn but room reflections are added into the response. 

The frequency response for the RCF ND950 on the 1723 horn is shown below. We see a similar response with a slightly shallower falling response this time starting at 1kHz. We see the typical diaphragm breakup starting to occur at 8kHz. 

Next I look at the dual SB Audience 65CDN-T on Sabourin Horn No.1300 which has a 300Hz cutoff frequency. We see good loading down to 300Hz and a level response out to 6kHz where we see a shallow 6dB/octave falling response above this. We see diaphragm breakup starting at around 14kHz, a virtue of the smaller 2.50" diaphragms. 

For interest's sake we can look at the impedance curves for each. Starting with the Celestion Axi2050, we see a FS of 242Hz and Re of 6.12ohm. The 'Q' of the fundamental resonance is quite high at 80ohms.

Below we see the impedance sweep for the RCF ND950 with the FS at 358Hz which is well above the cuttoff frequency of the 1723 horn (200Hz). Re is similar to the Axi at 6.4ohms (see bottom of graph).

This is a complete side note, but I removed the rear cover on the RCF just temporarily to see how this would affect the impedance sweep.

Below is the impedance sweep of the RCF ND950 with the rear cover removed. I secured the voice coil using the existing screws along with rubber coated flat washers. We can see that the FS drops from 358Hz down to 221Hz, even lower the Axi2050! 

 I then decided to measure the frequency response with the rear cover removed just to see how things would change. Below is the frequency response of the RCF ND950 with the rear cover removed (open back). 

We can see with the rear cover removed (red) we lose a little bit of output at 400Hz but a little more output at 200Hz. If I had to chose between the two for crossover development I would chose the open back. 

Below is the impedance sweep for the Dual 65CDN-T Horn 1300 combo. We see a FS centered at 279Hz with an Re of 3.53ohm since I have the two drivers wired in parallel. There is also virtually no inductive rise in the upper treble. 

Back to the study in question. Burst decay is shown below for the Celestion Axi2050. The resonances shown in the 500Hz to 1kHz region are reflections in my room so they can be ignored. I wanted to mainly focus on the area above 1kHz. The falling response somewhat masks low level noise in the 2kHz region. But it looks relatively clean in the driver's pass band from 200Hz to about 4kHz depending on application. 

Below is the burst decay for the RCF ND950. The region from 1kHz to 8kHz seems similarly clean to the Axi2050. This is no surprise since both drivers exhibit the same frequency response smoothness. 

Below is the burst decay for the dual 65CDN-T + 1300 combo. Things are a little less clean than the previous two. We do see the resonance behavior of the driver's FS coming in at 279Hz. 

The CSD plot for the celestion Axi2050 is shown below. Again, it's hard to tell much of anything from this plot due to the falling response. 

Below is the CSD for the RCF ND950. The result is similar to the Celestion however we do see the strong resonance breakup at 8kHz.

Below is the CSD plot for the dual 65CDN-T. We can see much more in this result since the response is quite level with no falling response. What we see is a very fast decay starting at 1kHz. 



I will test harmonic, intermodulation, and Gm at 85dB, 95dB and 105dB test signal levels. 

Below is the harmonic distortion for the Celestion Axi2050 at an 85dB test signal level at 1m. We see H2 at -70dB with H3 and H4 into the noise floor of the mic preamp. 


Below is the same as above but for the RCF ND950. We see the RCF struggle a bit in the 400Hz region. 

Out of curiosity I measured the RCF with the rear cover removed to see if this was related to the high FS with the cover on. Below is the harmonic distortion for the RCF ND950 running open back. We see an 8dB improvement distortion in the 400Hz region with the cover removed. Distortion through the rest of the bandwidth is similar to the Celestion. 

Harmonic distortion at 85dB for the Dual 65CDN-T is shown below. H2 is -75dB. This is a full 15dB better than the open back RCF and 4dB better than the Celestion. However there is more to discuss when we get to the 95dB test signal. 

Raising the test SPL to 95dB for the Celestion Axi2050 is shown below. We see H2 rise from -70db to -60dB in a linear fashion to the test signal increase. H3 and H4 approach the mic preamp noise floor. 

Doing the same on the RCF ND950 closed back is shown below. Distortion remains the same at the higher test SPL for the lower midrange, but it's still a bit higher distortion than the Celestion for the same test. 

Removing the rear cover is shown below. We see improvements mainly to H3 and H4, with H2 basically unchanged. So far we are not seeing any disadvantages to removing the rear cover. 

Looking at the dual 65CDN-T setup for the same 95dB test signal is shown below. H2 beats out the Celestion at -72dB and H3 and H4 is into the preamp noise floor. I should note however there there is definitely a low distortion null in this frequency region which doesn't represent the rest of the spectrum. However even if we look at the the region around 1.5kHz and compare the celestion against the dual 65CDN-T they are roughly the same at H2 -58dB and -56dB respectively. 

Below is the harmonic distortion for the Celestion at 105dB SPL 1m. We see H2 reduce again to -54dB (from -60dB). Even in the midrange at 1.5kHz we see harmonic drop to -44dB. So we are seeing lots of H2 comparatively speaking.

Shown below is the harmonic distortion for the RCF ND950 at 105dB SPL. We see H2 increase from -52dB to -46dB.  H3 improved slightly from -67 to -69. So we can see that as we increase the test SPL the RCF closes in on the celestion's distortion performance but doesn't quite match it. 

Removing the rear cover on the RCF we see distortion get a little worse than before by about 3dB. So keep the cover on if playing at high SPL. 

Showing the same but for the dual 65CDN-T is shown below. With the 105dB test signal we see H2 still pegged very low at -62dB. For reference the Celestion is -54dB for the same test SPL. So we've beat the Celestion by 8dB for this very high test SPL of 105dB. 

Conclusions On Harmonic Distortion 

When looking at the three drivers we see the dual 65CDN-T setup exceeding the Celestion's distortion performance by a wide margin of 8dB when looking at the region around 500Hz. The RCF trails third place by another 8dB.  The added advantage of the dual 65CDN-T is that it has wide bandwidth right up to 14kHz. We will see more of this come to light in the next part of the test... 

Intermodulation Distortion 

Below is the multitone multiband test tone signal ranging from 500Hz to 20kHz for the Celestion Axi2050. Distortion is very low in the 500Hz region showing a dynamic range of 83dB. However once we get past 1kHz we see IMD reduce to -70dB.  By 5kHz the dynamic range is -60dB. 

The same test is shown for the RCF ND950 below. Distortion performance is actually quite similar to the Celestion providing -82dB in the 500Hz region and increasing to -73dB for 1kHz, and -62dB for 5kHz. 


Removing the rear cover from the compression driver provides no change in the distortion performance, so we will leave this configuration out for the rest of the testing.

Looking at the dual 65CDN-T setup we see -80dB for 500Hz with the same result at 1kHz. So we don't see distortion rise until 2kHz increasing to -70dB.  At 5kHz we don't see distortion rise like the other two drivers, staying low at -70dB. 

If we increase the test SPL to 95dB we see the Celestion at 1kHz region lose 10dB of dynamic range to -60dB. We see the 5kHz region reduced to -50dB.

If we look at the the RCF we see -75dB for the 500Hz region reducing to -65dB for above 1kHz. 

The dual 65CDN-T shows -70dB for the 500Hz region increasing to -60dB above 2kHz. 


Increasing the test SPL to 105dB for the Celestion Axi2050 is shown below. We see -65dB for the 500Hz region increasing to -50dB for above 1kHz reducing to -40dB at 5kHz.

Looking at the RCF ND950 we see -65dB for the 500Hz region and -56dB for above 1kHz. 

Looking at the dual 65CDN-T we see -60dB for the 500Hz region increasing to -50dB above 2kHz. 

Gedlee Distortion Metric (Gm)

Using the Vertins Multi Instrument Tool Software I tested Gm using a 500Hz test tone for 85dB, 95dB, and 105dB. I've summarized the respective Gm values against the traditional THD figures for comparison. 


Conclusion on Gm Metric 

Generally speaking we see an inverse correlation between GM and THD-A. In other words as we increase the test signal SPL we see Gm decline (improve) and THD increase (worsen). 

With regards to the specific driver performance, we see the dual 65CDN-T configuration outperform both the RCF ND950 and Celestion Axi2050 by a wide margin when looking at Gm. 

Final Conclusions 

When looking at intermodulation distortion and Gedlee Metric (Gm) the dual 65CDN-T comes out as the winner with the RCF ND950 in second place and the Celestion Axi2050 in third place. However if looking at only harmonic distortion the Celestion comes in first place with the dual 65CDN-T coming in second and the RCF ND950 in third place. Since IMD and Gm is a more credible metric, I can conclude that the dual 65CDN-T offers the best performance with the added benefit of wide bandwidth out to 14kHz and wide off-axis coverage. 

Further Remarks 

This is the first time attempting a dual driver solution using the Y-adapter.

I had received feedback from my Facebook group that Y-adapters typically introduce problems and are generally not worth the trouble. However the Y-adapters available are very basic and crude attempts in terms of actual geometry. They appear to be severely compromised with many design flaws. I've made every attempt to create a perfectly smooth transition in my adapter design allowing zero chance for acoustical reflections. This seems to have worked. Based on the success of Y-adapter 2016-01-001 used in this test, I will be looking at possibly using this Y-adapter in my biradial horns. The ES-290 biradial (No.1670) could potentially benefit from this type of an adapter which has a smoother frequency response than Horn 1300. 

In the context of home hifi, we rarely see SPL levels beyond 95dB and it's much more common to listen at levels closer to 75dB. In this context I wanted to show the IMD performance of the dual 65CDN-T at a test signal SPL of only 75dB. 

We can see that IMD is at least -70dB or lower when listing at a moderate listening level. 


Even if we change the test and introduce test tones down to 200Hz we still see distortion remain low. 





Back to blog