Sound on Sound has a great article on Read more →
Leading up to Christmas last year my eldest daughter was putting together her wish list and asking me if I could get her a bluetooth speaker system. Looking at the prices and the quality of the devices she was suggesting I said I bet I could make something much better for a similar amount of money to which she happily agreed. I’d inadvertently sign myself up to building her a speaker system for Christmas and to be fair on her younger sister, I needed to make two.
At the outset she expressed a desire for the system to support the same functionality as the device she was initially wanting me to buy. That mean’t blue tooth connectivity with volume, play / pause, track forward and track previous functionality. All those features are easily supported with the RN52 bluetooth audio module from Microchip, however, as the device is of small form factor and pin pitch of 1.2mm, it is not well suited to hand soldering. Typically it would be soldered on a surface mount PCB using pick and place robots and a soldering oven which is certainly something not available to me. Fortunately though, a hobbyist electronics outfit called Sparkfun sells a break out board which brings all the device pins out onto a 0.1inch pitch hand solderable board. Problem solved for the bluetooth connectivity.
Next comes the speaker choices. To keep costs to a minimum I made a decision early on to only look at full range drivers to eliminate the added complexity of crossovers and extra amplifier modules. Looking through the catalogue of available devices available at Parts Express I quickly found out that there is a huge range of possibilities of varying quality. Ultimately my aim was to build a system that has passable high frequency performance but with a better than average bass response to make it a stand out in comparison to the systems my daughter was looking at, systems which were costing around $200-$300. Because of the manufacturers reputation, my previous experience with the quality of there drivers and the manufacturers measured response I settled upon a peerless-830987 3 inch anodised aluminium full range driver.
Again to keep costs to a minimum, for the choice of amplification I restricted myself to a range of pre-assembled amplifier modules with power outputs in the range of 10-20 Watts per channel. For this task I settled upon the Sure Electronics AA-AB32155 amplifier module that delivers 15 Watts per channel into a 4 Ohm load. The whole system would be powered by a laptop style switch mode power supply module.
I wanted to obtain the best possible bass extension from the driver in a box of relatively small volume. Naturally I chose a vented enclosure as this would enable me to obtain a useful response below 80Hz. Doing some manual iterative vented enclosure designing using R scripts based on those I used in developing my full range monitors, I settled upon a box volume of 3 litres and a box frequency of 65Hz. This combination gives a relatively flat bass response that is -3dB down at a quite remarkable 54Hz. The modeled response is shown below.
Modeled Woofer Frequency Response
Modeled Woofer Impedance
Although the response is not maximally flat it only has a modest positive deviation of +2dB at around 120Hz. Whatever design parameters we choose we cannot remove this deviation as it results from the driver having a high Qt of 0.94. I reasoned that whilst it will introduce a bass colouration, the degree is modest and likely to be welcomed by my daughter as “louder bass”.
Before embarking upon building the speaker cabinet proper I wanted to test the design and have a means of easily testing the audible performance of different left-right driver separation and angles so I built two 3 litre enclosures with the correct box tuning. These cabinets did indeed confirm the bass extension of the drivers which we found remarkable given there size, however, I could readily hear a mid-range stridency at around 2kHz which my ears found distressing to listen to. Close inspection of the manufacturers measure response confirms the stridency in that region as can be seen below.
Peerless 830987 Driver Response
Although my daughters didn’t comment on it, I certainly quickly came to the conclusion that to obtain the best possible performance from these good drivers I would need to equalise this stridency out. Going from the manufacturers measured response graph and listening tests using the test cabinets and amplifier module being driven by Har-Bal and some heavy rock / grunge music, I came to the conclusion that I needed a notch filter with a centre frequency of 2kHz, a Q of 5 and an attenuation of -6dB. Listening to a filter of those characteristics realised through Har-Bal confirmed it to be sonically pleasing and the best possible balance for a single notch response filter.
In terms of driver arrangements, I came to the conclusion that the most satifying sound for a combined system whereby both drivers share the one physical cabinet was to have them angled outward. By doing so the interference pattern between the drivers would be reduced and the room reflected sound would have a definite stereo aspect to it, more-so than if the drivers shared the same plane baffle.
To obtain the desired performance we need to equalise the driver. As neither the RN52 or the amplifier module have that capability we need to realise it with extra electronics. Ideally, to keep added noise to a minimum it would have been great to be able to realise it with digital filter but that would add both expense and complexity to the small scale project so my only alternative is analog electronics. This is easily realised with a potential divider circuit in which the grounding arm is a series LCR circuit with the L / inductor component realised using a gyrator circuit (active electronics that makes a capacitor look like and inductor). The NGSPICE simulated response of the EQ is show below.
Simulated Equaliser Response
We also need a regulated 3.3 volt supply to power the bluetooth receiver electronics and to improve the utility of the device I also decided to add a 5 volt supply powering a USB port for use exclusively as a charging / powering port for hand held devices / tablets etc. The 5 volt supply would also provide a bias level for the analog electronics which would be powered off the 12 volt supply supplying the amplifier module.
The other aspect I discovered on the run was that I would need a differential to single-ended amplifier to interface the RN52 to the analog electronics. Initially I was thinking I could use it single-ended but this has two complications. Firstly, the RN52 ground is heavily contaminated with digital noise so trying to run it single ended results in quite audible digital noise contamination. Secondly, when the RN52 goes for sleep state to pairable and from paired to connected the DC output level from the output line goes from 0 to 1.7 volts or so which results in a massive glitch coming from the speakers. The only way of avoiding this is by using the differential outputs and converting them to single ended for the sake of the remaining analog electronics.
The electronics design I came up with is shown below. Although it shows a de-click circuit for the loudspeaker I ultimately didn’t include it as I found the level of turn-on click delivered to the speakers was too small to justify the added complexity.
Bluetooth Module Rx’er
Input Selection, Equaliser and Power Amp
I had quite ambitious goals in making interesting and stylistic cabinets for these two bluetooth speakers (one for each daughter). Most problematic being the use of angles other than right angles. Anyone with any woodworking experience will readily tell you how much more complicated everything gets when you are no longer working in right angles exclusively. That and doing things by the seat of my pants really did push my capabilities to their current limits.
Even after completing the basic cabinets I didn’t have a clear idea of how I was going to be incorporating the electronics nor if I’d have enough space to fit it all in. I knew how big the amp module was, how big my switch panel was going to be, but beyond that it was just working it out as you go, which in hindsight is probably not the best way to handle it. I did attempt to figure out the dimension of the top and bottom cabinet pieces via plans but on finding that they didn’t work (ie. were wrong) I just winged it by eye from the other cut pieces and dispensed with the plans.
The electronics too posed a number of challenges along the way. The first being the LM317 not regulating the 3.3volts required which, as I found out, was because it has a minimum load requirement that I wasn’t satisfying. Following that issue I discovered that using the output of the RN52 single ended was not going to fly, the output level was too low and the bluetooth control buttons weren’t functioning. The later points I discovered were fixable by customising the RN52 which is done through programming it via commands issued through a serial interface. To be able to do that I need to construct a TTL to RS232 level converter so I could program it with my laptop, although it didn’t end there. When attempting to program it I discovered it didn’t support the functionality the documentation said it would. Why? Because the firmware was out of date. Then I had to construct the DFU interface and download and flash the most up to date firmware for the device so that it finally it supported what it said it would.
Finally, apart from a number of missed steps in incorrect wiring, I had major issues solving ground noise problems which I never did completely find an answer to. Ground noise issues can be extremely mysterious and this was no exception. I managed to fix the bluetooth path so it only had a minor mains hum but intriguingly when connecting a double insulated CD player to the line input it immediately gave a loud mains hum and plugging in a iPod mini would produce the same hum if you touched the device. The odd thing about the whole behaviour was that if I grounded it to my Oscilloscope (which is connected to mains earth) all the ground noise problems would disappear. As I don’t have a great ideal of why the floating ground is causing these issues I just hacked a solution by providing a mains earth thus converting the floating earth – double insulated power supply into a mains earthed (but still double insulated) one. Looks weird but it works a treat.
Cabinet from the front with Speakers in place and Covers removed but no Electronics.
Cabinet from the rear showing the Electronics Bay.
Electronics Carriage which slides into the Electronics Bay.
Electronics carriage front panel.
Electronics carriage rear panel.
Completed and Assembled Speaker System.
The Proof in the Pudding
Ultimately the question comes as to how well do they perform. Honestly, I was completely gobsmacked as to how good they sound! So much so that when I finally completed one and got to listen to it at around 11:00 at night I spend the next two hours listening to everything and anything I could find. More than anything else, the bass extension just surprises me, that and the tightness of it. The clarity of the mid-range is exceptional and in perfect balance with the bass and although the top end is not perfect the nature of the colouration is such that you don’t really register it as a problem. If anything, it gives an exaggerated presence to transient material. I am so pleased with the way they sound I’d like to keep one for myself but alas, I can’t, because they are now my daughters speakers and I’m too tired from the experience to want to build another one right now. Maybe some day in the future but not now! I’ll just borrow one of there’s. Seriously, if I had to put a dollar value on how they sound I’d be thinking something around the $800 mark or so. They certainly blow anything in the $300 market out of the water no holes barred. The angled arrangement of the drivers really works well too, giving a surprisingly good sense of stereo from drivers close together and interestingly enough, I find the most pleasant sound is to be found off axis on either side. Whilst the centre line provides a fine performance there is audible interference happening which you just don’t hear slightly off axis (15 degrees or so). I really love these!
On a closing comment after many more hours of listening to them I’d would add that these speakers are more “modern popular music” friendly than my reference monitors. Many recordings these days being recorded in home studios often have an overbearing bass end brought about by mixing and mastering on small speakers with a limited low end output. It just so happens that the bass response of these speakers actually matches the “average” home studio monitor pretty well so consequently they are much more compatible with these recordings than my monitors. These make what I can’t bear listening too on my monitors perfectly listenable and the key reason I’d love to keep one for myself!
I have received some detailed criticism on my loudspeaker design because of my choice of a vented enclosure. The basis of the criticism is that vented enclosures can have poor transient response if there is insufficient damping of the port resonance. Even with a well designed vented enclosure the transient response of a comparable sealed enclosure will be better because it has a second order high pass response rather than a 4th order one in the case of a vented enclosure.
However, to make a comparable sealed enclosure is considerably more expensive as it will involve larger drivers (need to lower free air resonance) and considerably larger cabinets. If I chose to make a sealed enclosure with the same drivers the low frequency cutoff of the loudspeaker system would be about 50Hz and the box volume of around 50 litres. It is slightly smaller but the cutoff frequency is nearly twice as high. In short, it is a decision of pragmatism to be able to obtain the best performance at the lowest cost that I chose a vented enclosure.
Validating Transient Response
To illustrate that the loudspeaker design does not behave as a “boom box” or a one note bass system and to also demonstrate the transient behaviour of the entire system I conducted some tone burst tests across the entire spectrum. These tone burst tests are as shown below. The lower trace is the tone burst input and the upper trace the acoustic output.
The 30Hz case represents the worst case scenario for the vented enclosure as the box frequency of the design is 30Hz. Whilst by no means perfect, the carry over in the absence of the excitation is quite well controlled and certainly not strongly resonant. It is 20dB down within a cycle.
The 60Hz case is similar to the 30Hz case in terms of carry over but for this frequency most of the acoustic output comes from the woofer and not the vent itself.
The 100Hz tone burst response is very good and shows very little resonance. The 300Hz and 600Hz responses show a similar response but do exhibit what looks like some degree of cabinet / basket resonance although it is 20dB or more down. Perhaps if I had bothered to internally line the cabinets with some sort of viscous damping material (bitumen for example) these responses may have been better.
The 1kHz case is perhaps the worst response in the entire set of measurements though it probably isn’t entirely surprising that that is the case. 1KHz is the crossover frequency so the third order cutoff of the crossover will add some ringing and the response could also possibly be confused by having the acoustic output being generated by the tweeter and woofer-mid in equal proportion, as there are two propagation paths for sound to the microphone with possibly different time delays (if I didn’t position the microphone at the exact acoustic middle of the two drivers). It may well be that the majority of the carry over in this case is tied up with output from the woofer-mid, as it is operating outside the piston frequency range for that driver.
The 3kHz, 6kHz and 10kHz responses are very good and indicate that the tweeter is of high quality.
Although a vented enclosure is used the vent is by no means a monotone device or strongly resonant. The low frequency tone burst response below 100Hz whilst not perfect, are in no way bad. I would argue that given the design constraints they are quite good.
The tone burst response in the low – mid range (100Hz, 300Hz and 600Hz) are well controlled but do show some low level structural resonances. The degree to which they are are function of the driver or the cabinet is not know and would take considerable effort to isolate.
The crossover region between the woofer-mid and the tweeter of 1kHz shows the worst tone burst behaviour out of the entire set, though not entirely unexpected. The crossover will add some ringing by virtue of being a 3rd order crossover, but I believe the majority of the carry over is due to modal resonances in the woofer-mid driver cone.
The tone burst responses above 1kHz are all very good indicating that the tweeter has a high quality response.
Although not perfect, taken overall, the responses a pretty good and tally with what I hear when listening to them. I don’t see any significant evidence of the speaker transient response being unduly compromised by the choice of vented enclosure.
As promised much too long ago, here are the detailed plans and bill of materials to construct my new loudspeaker system.
The PCB (printed circuit board) layout bitmaps are (or at least should be) correctly scaled so that if you print them out at 1:1 scale you should have the true size of the boards. The mid range/tweeter crossover page size is A3 and the woofer crossover page size is A4.
Far from being printed, I actually built them in a very crude manner by simply taking a single sided PCB blank, marking the cut pattern on it with a marker pen and cutting (with a small angle grinder) rather than etching, the board and drilling the holes. The grey colour corresponds to copper that needs to be etched or cut, whichever way you choose.
The large holes in the PCB’s along side each inductor are for secure mounting purposes which I achieved through the use of nylon cable ties to hold them securely in place (they are pretty heavy owing to the cores). After the fact I discovered that it is important to secure the components via other means to dampen resonances. The approach I took was to use silicone rubber to secure the components to the boards to reduce component vibration. Similarly, when mounting the crossover boards inside the cabinet care must also be taken to ensure that the board vibrations are damped out. For that I simply used some of the packing foam the inductors came encased in.
Final Crossover Design Schematic Diagram
Mid-range/Tweeter Crossover PCB design – (15 by 30 cm in size).
Woofer Crossover PCB design – (10 by 20 cm in size).
The cabinets should be constructed with 18 mm MDF or something similarly heavy and robust. You may choose to build it in a different shape as shown as long as the volumes are preserved and the driver layout is as shown with the mid-range unit mounted above and the woofer mounted below the tweeter.
Although the plan does not show it, I recommend adding extra bracing behind the front facia between the holes cut for the tweeter and mid-range units as the mid-range can give rise to significant cabinet colouration arising from structural vibrations. I found this out after the fact, initially assuming it would be problematic so couldn’t easily retrofit such bracing.
The woofer- mid unit also requires a stand-off to time align it to the tweeter for which plans are shown below. I constructed this from two sheets of 12mm MDF cut to shape and glued together, then painted to beautify. Note that the cusp on the stand-off should point to the centre of the tweeter when the drivers are mounted.
During assembly gaskets are required for the tweeter and woofer-mid range drivers, the standoff and the vent (if you choose to use the “Precision Port” that I used). For this I used un-polished leather which works very well as a gasket material and is readily available from leather craft stores.
The port length should be adjusted to obtain the required 30Hz box frequency. If you use the “Precision Port” in the bill of materials then this is obtained with a pipe length of 30 mm joining the two flanges.
How you choose to mount the PCB’s I leave up to you. In my case I screwed them to the back panel of the cabinet just behind the woofer. making them accessible and constructable without the need of an access port. That was also the reason for two PCB’s because the boards needed to be small enough to be able to maneuver into the cabinet through the woofer port.
The woofer-mid-range enclosure should be filled with accetate wool / Insulation. The stuff I used I got from a haberdashery store which was sold as cushion stuffing material. You could equally well use material meant for thermal insulation. Most of the woofer cabinet should be free of acoustic insulation to obtain the right box Q. In my case I filled the bottom 1/4 of the cabinet with accetate wool which appears to give a box Q pretty close to the design ideal.
Basic Cabinet Design
Woofer-Mid Standoff Design
Bill of Materials
2 of Peerless SDS-830656 5 1/4″ Woofer-Midrange
2 of Peerless SLS-830667 8″ Sub-Woofer
2 of Precision Port 3″ flared port tube kit
2 of Gold Banana 5-way recessed speaker terminal round
2 of Vifa XT25TG-30-04 Dual Cone Super Tweeter
2 of 1 uF 100V non-polarised electrolytic capacitor [ C16 ]
2 of 4.7 uF Jantzen 400V crosscap capacitor [ C4 ]
2 of 10 uF 100V non-polarised electrolytic capacitor [ C7 ]
2 of 18 uF Jantzen 400V crosscap capacitor [ C14 ]
2 of 22 uF Jantzen 400V crosscap capacitor [ C3 ]
2 of 27 uF 100V non-polarised electrolytic capacitor [ C11 ]
4 of 33 uF 100V non-polarised electrolytic capacitor [ C8, C18 ]
2 of 47 uF Jantzen 400V crosscap capacitor [ C2 ]
2 of 56 uF Jantzen 400V crosscap capacitor [ C15 ]
4 of 80 uF 100V non-polarised electrolytic capacitor [ C5, C6 ]
8 of 200 uF 100V non-polarised electrolytic capacitor [ C1, C9, C12, C17 ]
2 of 250 uF 100V non-polarised electrolytic capacitor [ C13 ]
2 of 400 uF 100V non-polarised electrolytic capacitor [ C10 ]
2 of 0.05 mH Jantzen 16AWG air cored 0.06 ohm DCR inductor [ L9 ]
2 of 1.5 mH Jantzen 15AWG P cored 0.14 ohm DCR inductor [ L1 ]
2 of 0.45 mH Jantzen 16AWG air cored 0.25 ohm DCR inductor [ L2 ]
4 of 0.56 mH Jantzen 15AWG P cored 0.08 ohm DCR inductor [ L10, L11 ]
4 of 3.3 mH Jantzen 18AWG P cored 0.35 ohm DCR inductor [ L3, L4 ]
2 of 5.6 mH Jantzen 18AWG P cored 0.45 ohm DCR inductor [ L7 ]
2 of 12 mH Jantzen 18AWG P cored 0.57 ohm DCR inductor [ L5 ]
2 of 18 mH Jantzen 18AWG P cored 0.83 ohm DCR inductor [ L8 ]
2 of 22 mH Jantzen 18AWG P cored 0.96 ohm DCR inductor [ L6 ]
6 of 0.33 ohm wirewound 5W resistor [ R1, R2, R11 ]
2 of 2.2 ohm wirewound 5W resistor [ R4 ]
4 of 2.7 ohm wirewound 5W resistor [ R5, R6 ]
4 of 3.3 ohm wirewound 5W resistor [ R9, R10 ]
2 of 6.8 ohm wirewound 10W resistor [ R3 ]
4 of 8.2 ohm wirewound 10W resistor [ R7, R8 ]
You should be able to purchase all the parts required through Parts-Express.
Having now had my new speakers for 9 months or more and having had many hours listening with them, I can say with complete honesty that I find them a pleasurable listening experience. It is easy to forget that the sound is emanating from loudspeakers and to feel immersed in the experience. I’ve yet to hear with any certainty a audible problem with the sound reproduction of these loudspeakers. If I ever hear disturbing things it is invariably due to problems in the recording itself.
What really impresses me about these speakers are the bass extension, the reproduction of voice and the well balanced sound in the far field. It genuinely fills the living room and open plan kitchen with well balanced sound even though the speakers are not in the living room, but rather situated in my adjacent small “music room”. The bass extension has, on many occasions, indicated low frequency noise problems in recordings that simply aren’t audible through my old speakers.
I have said it before and I shall say it again with confidence. Anyone who has the courage to build this loudspeaker system will be impressed by their performance but I can well understand that such an ambitious project may well be beyond most, if for no other reason than the cost of the components required to build them. The drivers aren’t expensive but the number of components in the crossover escalates the cost of the system, so much so that the crossovers cost nearly twice as much as the drivers! However, skimping on the crossover design will only come at the cost of performance and that really wasn’t my aim so I’d argue the added cost is more than worth it. My system cost me about $1200 and my labour to construct. There is absolutely no way I could get anything approaching this level of performance for that sum of money and I’d argue that I’d be spending in excess of $5000 on a commercial system to obtain similar capabilities.
All in all I thoroughly enjoy listening to my new speakers and have absolutely no desire to replace them any time soon, perhaps never.
It must have been sometime last year during my epic loudspeaker design project (in testing phase) that I noticed, to my shock, that when I played “Something For Kate’s” track “California” (from there album “Desert Lights”) using my Nexus 7 tablet as source, I enjoyed it a whole lot more than if I listened to the CD (from which the the tracks were downloaded) directly.
I remember at the time it somewhat stunned me, but I didn’t think much more of it until today, when it finally struck me as to why that might be. As with much over limited material these days, clipping is a pretty common place event and that particular CD is no exception.
The presence of clipping is the key to answering the question. Clipping introduces wide-band noise into the signal chain, which lossy compression schemes like mp3, or ogg vorbis for that matter, simply can’t encode accurately, so the upshot is that the information lost out of lossy encoding the clipping ends up smoothing it out and making it sound less distorted. You can clearly see this at work in the time line of the “California” track in the samples I show below.
Clipped segment in “California” track as is on CD
The same clipped segment in “California” track after being converted to mp3 format
Looking at CD case (click on it to get a bigger view) you’ll see a clear and extreme clipping event in the content, which is evident from the ruler flat top of the signal. Now look at the same event after being converted to mp3 and you’ll see that although it still looks like clipping, it is no longer ruler flat and has sharp edges smoothed out. It is the sharp edges that give rise to the wideband noise that makes the clipping sound so obviously bad.
The general finding following from this behaviour is that any content that conforms to loudness wars playback levels will generally always sound superior as an mp3 rather than the unadulterated and more accurate CD source, not because the CD format is bad, but because the mastering is rubbish and the CD shows it up as such, whilst the lower Fi mp3 has a harder time of showing up the defects.
Given the entrenched nature of the loudness wars mentality, the only reasons you should consider actually buying a CD version of commercial music is to have a backup and to have the enjoyment of looking through the liner notes and artwork. You should probably set about transferring it to your mobile phone / tablet to actually play the content and not bother with the CD at all, other than for a definitive backup of the material.
If the music business wanted to continue selling CD’s more than downloads, the wholesale participation in the loudness wars did them no favours. That participation has essentially destroyed the aspiration of Hi-Fidelity to such an extent that you would be mad to even consider it if you are into popular contemporary music. Perhaps if you are into classical music but even there I’ve notice with horror how loudness levels are being pushed higher.
Perhaps the only reason to consider a Hi-Fi purchase these days is for home cinema reasoning where quality recording technique is still practiced and appears to be the entrenched (thankfully) approach.
In my previous frequency response measurements posted for the unfinished prototypes I inadvertently had the woofer wired in the wrong phase leading to a deep notch in the crossover point between the woofer and mid-range drivers. Having now fully completed both loudspeakers and rectified the wiring error I thought it appropriate to post a new set of performances plots to reflect the true nature of the loudspeaker design.
Starting with the near field on tweeter-mid range mid line frequency response we see a nominally flat response with no easily discernible crossover region problems. The notching at 120Hz in the previous test is not present here.
On tweeter-mid range mid-line frequency response at 50cm
At 45 degrees off mid-line in the horizontal plane the response also shows good crossover behaviour with the expected drooping response at the high frequency end due to the polar response of the tweeter.
45 degrees off tweeter-mid range mid-line frequency response at 50cm in horizontal plane
At +45 degrees off mid-line in the vertical plane there is a peaking around 1kHz and some dipping in the 2-3kHz region which is largely consistent with modeling.
+45 degrees off tweeter-mid range mid-line frequency response at 50cm in vertical plane
Similarly, at -45 degrees off mid-line in the vertical plane there is notching around 1kHz which is also largely consistent with modeling.
-45 degrees off tweeter-mid range mid-line frequency response at 50cm in vertical plane
Moving further afield, the response of the left channel in my normal listening position is nominally flat from around 20Hz to 20kHz with some deep notching in the 30Hz to 200Hz region. These notches are a result of the standing wave patterns in my listening room which is, by typical monitoring standards, an acoustically small room. It is pentagonal in shape with a side length of around 2.7m so is a very modest room size, hence the problem with low frequency standing waves.
Frequency response at listening position for left channel approximately 1.1m from loudspeaker
For the right channel the result is similar though the notching at the 30Hz to 200Hz region is distinctly different suggesting that my room is not acoustically symmetrical or at least my speaker placement. However, in both cases it is pretty clear that the response extends down to 27Hz as the design figure suggested.
Frequency response at listening position for right channel approximately 1.1m from loudspeaker
Going slightly further afield to a position just behind my listening position we see the bass lift in both left and right channel responses owing to the microphone moving into the sphere of influence of the room modes (ie. closer to the wall) whereas my listening position is close to the mode anti-node, hence the deep notching and controlled bass output.
Frequency response behind listening position for left channel approximately 1.5m from loudspeaker
Here we see that the influence of the room extends the useful bass output right down to 20Hz making it clear that it covers the normally accepted complete audio spectrum. No need for sub-woofers here!
Frequency response behind listening position for right channel approximately 1.5m from loudspeaker
Finally, I’ve noted on many occasions through my normal breakfast and dinner preparation listening that takes place in my adjacent open plan kitchen, that the sound I hear there is remarkably clear and uniform despite not being in any way optimal as far as listening positions go and significantly clearer than with my previous loudspeakers. In fact the sound simply propagates through the open doorway to my normal monitoring room so you would hardly expect high fidelity but to confirm the impression I did some measurements of the sound picked up in that room without a direct line of sight to the loudspeakers. The results speaker for themselves showing a response that is essentially nominally flat from 20Hz up to around 7-8KHz where upon it begins drooping off! Quite an impressive result I thought and perhaps a vindication to the design approach and the attention played to the vertical polar response.
Frequency response for measured for left channel in adjacent room to my listening room approximately 6m from speaker without a direct line of sight
Frequency response for measured for right channel in adjacent room to my listening room approximately 6m from speaker without a direct line of sight
In conclusion, I think it is clear that this design works incredibly well and is worthy of monitor status. Certainly it is more than adequate for my needs and probably out performs speakers costing much much more. The only real issue I have to now deal with is the problem of deep notching caused by the standing waves in my acoustically very small room. This too, I believe I have a solution for in the form of echo cancellation equalisation techniques but I’ve yet to implement the algorithm to design the EQ or purchased the DSP hardware to implement the EQ in, though I have done some rudimentary tests using a DAW to verify that it is at least technically possible but that is another project for another time.
Construction to Completion
Making the loudspeaker cabinets “unfinished” was easy, perhaps two days work. Actually giving them a stylish finish is much more time consuming.
The MDF cabinets will have a real timber veneer applied to it. Actually, I’ve already done one cabinet.
Application of the veneer is a tedious and time consuming business requiring careful application of the veneer with contact adhesive and then trimming to the cabinet size. This was made worse for me by the choice of one of the veneers (iron wood) which although attractive, is very hard and brittle, so not easily trimmed to fit without splitting it. As such the side panels, which are mostly covered with iron wood are only covered to within about 5 mm of the edges so that I did not need to trim it, but it meant adding an extra thin piece on either side to take it to the edge, using a different timber that is more forgiving to trimming. The final finished and vanished cabinet is show below.
The crossover boards are mounted on the inside rear panel.
All assembled the cabinet looks quite attractive. Note the flange setting the mid-range forward to time align it with the tweeter. The speaker sitting next to it is my previous model I built back in the 90’s.
Now compare it to the look of the yet to be finished prototype. Clearly far less attractive, but it has taken the best part of 7 days work to do the finishing work so I have plenty of labour still to come before I have a matching pair.
Although I had a tasting of what was to come when listening to my prototype as a single speaker along side one of my old ones, how they would sound as a pair was hard to envisage. Now that I have a working pair I don’t need to guess.
The sound is truly sublime. Imagine is accurate and it is really difficult to fix on a “problem” in the sound. There just doesn’t seem to be any obvious audible weaknesses in the sound. Likewise, there is absolutely no sense of crossover point weakness that I can hear in my old design and in other loudspeakers I had a chance to listen to. The sound is, in short, very accurate.
Highs are crystal clear but not overblown, mids are sharp and well defined, bass is extended and punchy and it all sounds cohesive as a whole. Listening to quality classical recordings it is easy to forget the speakers and become totally immersed in the performance, which comes across with all the warmth and detail of a live performance. The bass extension is so good that it disturbingly shows up recorded sub-sonic noises that probably shouldn’t be in there (eg. wobbling mic stands) and when it is intended it is shuddering without loss of control.
In fact, if I were to give praise for any one thing it is the low mid and bass reproduction which is seemless. So often big bass speakers have oomph but no tight punch, or conversely, plenty of punch but no oomph. This system has both. Perhaps most importantly, the reproduction of human voice is astonishingly clear, with a definition I don’t think I’ve ever experienced before, except in the case of the real thing.
All in all, they are a pleasure to listen to and well worth the effort of design and construction! Certainly, I think I can say with absolute certainty that anyone who has the courage to build this design won’t be disappointed with the sound quality that will result. Although not cheap (mine will have cost me around $1200 to build), I very much doubt you could find a comparable performance in a bought system for less than $4000 and that is probably conservative.
A final point on the measured responses in my previous post : the hole in the response around 100Hz is not a standing wave issue but a “constructor incompetence” issue. I wired the woofer up in the wrong phase which resulted in the notching. With the correct phasing the hole disappears and the response is nominally flat (excepting the peaks and troughs resulting from room reflections). When I have time I’ll re-do the results to show the real response of the system. I’ll also bundle together a bill of materials and all the design drawings needed to contruct the loudspeaker system. It may take a while though as I’ll be distracted by the sound of my new speakers, which I haven’t even finished yet!
The elaborate processes I’ve used in designing this loudspeaker system counts for nothing if not realised in a concrete form. Out of extreme curiousity as to whether this labour will pay any dividends, I worked frantically last weekend to build a box (less finishing details such as veneer and the like), then a late evening to assemble the drivers and crossovers into this one prototype so I could hear how it might perform.
Having only one meant listening to it when paired up with my previous design which fortuitously has about the same sensitivity. This gives rise to a rather skewed stereo image because their response differ but it does actually give a pretty clear assessment of how they compare one against the other.
At the outset it is clear my new design gazumps my long held previous speakers which I’ve loved for too many years. In fact, I loved them so much so that I never really felt compelled to get something better and really only took on this project as something to write about, but I’m not disappointed in my foray. The sound eminating from this design is entirely neutral, not having a colour of its own, but truly reflecting the colour of the recorded material, as it should. In contrast, my previous design has a mellow warm tone that verges on muddy with bass heavy material.
The thing that did concern me was having an excess of top end as I know from experience that too much tends to take all the depth perception out of sound reproduction, giving rise to something that is sharp and clear but one dimensional, however, this doesn’t happen and they reproduce depth well. The hemisphere of sound on the right side (the side my new design is currently sitting) sounds more detailed and space filling than what I hear on the left. The bass is also truly extended but controlled so although it is there down to 27 Hz it doesn’t try to dominate so the box design sounds essentially correct.
From a listening point of view I’m already sold on it but I did a few checks on the measured response to confirm to myself that I’m not imaging it and to perhaps figure out why they sound as good as they do. Below is a set of measure responses measured at a distance of around 50cm from the tweeter in my listening room at various angles off the central axis. You should ignore anything in the response below 20Hz as that will be inaccurate and the large peaks and troughs in the 20Hz-200Hz region are standing wave issues in my room, not the speaker response itself.
New Design Response : on tweeter axis at 50cm
The on axis response looks remarkably flat and extends down to the 27Hz region predicted by the modelling. There is no obvious problems with the mid-tweeter crossover point.
New Design Response : +45 degrees off tweeter axis at 50cm in the vertical plane
In the vertical plane at 45 degrees off axis (above the tweeter moving toward the ceiling) the response is similarly very uniform apart from the expected drop in sensitivity at the high frequency end. You would be hard pressed to pick a crossover point from the response.
New Design Response : -45 degrees off tweeter axis at 50cm in the vertical plane
In the vertical plane at -45 degrees off axis (below the tweeter moving toward the floor) the response isn’t so good, which is what modelling predicted, though it holds up reasonably well. The big dip at 150Hz is a standing wave issue and not a crossover issue.
New Design Response : +45 degrees off tweeter axis at 50cm in the horizontal plane
In the horizontal plane at 45 degrees off axis the response is almost as good as in the first vertical plane case, showing some peaking at 300Hz and 1.5kHz but again is pretty uniform and well behaved. All in all, it goes very close to behaving like a single driver rather than the three drivers it is composed of.
Now compare this to my old speakers to get a quality comparison. The on axis case is pretty uniform though you can see the origin of the warm sound in this design from the drooping top end. I adjusted the tweeter attenuation to taste and I think there are two reasons for it’s subdued nature, one being the resonance in the tweeter at 12kHz which makes it cut through quite a lot so I always found a flatter adjustment sounded too bright. The second issue is that the bass extends only to 40Hz and not the 27Hz of the new design. This meant that the flatter I’d made the top end the lighter the bass sounded and I preferred the bass cutting through a little more, even though it came at the expense of some tendency to muddiness.
Old Design Response : on tweeter axis at 50cm
Moving off axis we see problems creep in and out. There is definite mid-range weakness in the 1-3kHz region +45 degrees off axis in the vertical plane.
Old Design Response : +45 degrees off tweeter axis at 50cm in the vertical plane
In the other direction the story is much the same though possibly more so.
Old Design Response : -45 degrees off tweeter axis at 50cm in the vertical plane
In the horizontal plane the off axis response holds up better though there is some weakness around 4kHz.
Old Design Response : +45 degrees off tweeter axis at 50cm in the horizontal plane
Overall, it is clear that the difference in sound quality is not just down to the deliberately subdued response of my old design but also due to the less accomplished off axis response. It is not surprising that they should be inferior. There are multiple reasons. Firstly, the crossover frequency is 2kHz and not 1kHz meaning the effective driver separation is greater (because it is the separation in terms of wavelengths that counts) in the old design, secondly, I made no attempt to time align the tweeter and mid-range, and thirdly, the crossover was essentially hand optimised and not optimised through numerical minimisation as was here. The crossover is simpler too, omitting the phase equalisation stage that is included in this new design.
The Tweeter Sensitivity Mystery
This brings me to one final point about the design. The crossover design used in the final prototype differs from the one I originally published by the amount of padding applied to the tweeter. Using the padding in the original plans gave rise to a response way too bright.
The strange thing is that the padding was based on manufacturer supplied driver sensitivities so you would have though it should be close but I used about 6 dB more attenuation in the above as I had in the original design. Stranger too is the fact that I had the exact same issue in building my previous speaker system whereby building the crossover based on manufacturer supplied sensitivities led to a response way too bright. Why this is is still a mystery to me. You certainly can’t explain it through insertion loss from the crossovers as that figure is more like 1 dB and not 6 dB.
So if you ever attempt to design your own speaker system my advise to you is to base driver padding on what you hear and not what the manufacturers specs suggest.
As I alluded to if you happened to read (or at least skim) through the paper I wrote discussing loudspeaker theory, the tuning of a vented enclosure can be optimised by measuring the impedance of the woofer mounted in box with vent and stuffing in place. The box frequency is the frequency at which the minimum impedance occurs between the two impedance peaks and for our design should be 30Hz or there about. The level of mechanical damping in the system can be inferred by the height of the impedance peaks.
I’ve built a prototype cabinet out of 18mm MDF and mounted the woofer and vent and measured the resonant frequencies of the enclosure. Through a comparison with the theoretical I could see that my vent was too long owing to the resonances being too low so I adjust the length proportionately and got the near ideal tuning shown below.
The red plot shows the theoretical driver impedance and the green plot the measured impedance. The lower resonance peak pretty much coincides but the upper one is a little lower than theoretical. Plotting a second theoretical curve but with the resonant frequency of the driver shifted from 42Hz to 39Hz produces the blue plot which is almost identical to the measured, apart from the deviation at high frequency, which shows a considerably larger leakage inductance than claimed in the manufacturers specs. From this I’m guessing the deviation is basically due to the free air resonance of the driver being slightly lower than the manufacturers specs but well within reported tolerance.
The other thing to note is that the height of the impedance peaks between the theoretical and measured are almost the same meaning the actual box Q is about the same as the design figure. However, the blue plot has the box Q parameter set to 6 whereas the red plot has it set to 7 meaning the actual box Q looks closer to 6 than 7. This is interesting because the box isn’t lined with acoustic insulation. All it had in it at the time of measurement was about a 30cm deep layer of polyester pillow stuffing loosely packed down the bottom of the enclosure and the remainder is completely untreated. It should then be obvious that completely lining the box is probably a bad idea as it will lead to a lower box Q than we desire.
Having listened to a completed prototype I can say that having this level of stuffing does not present a problem in terms of box colouration. Without consciously considering it, this is unlikely to be an issue in this design because the standing wave and panel resonance frequencies are outside of the operating range of the driver which is cut at 150Hz. As such, I haven’t bothered treated the box with bracing for stiffening or viscous material lining the insider walls (such as bitumen) without audible detriment. If the woofer were crossed at 400Hz instead it would undoubtedly be a different matter.
Designing the Cabinet for a loudspeaker system is less involved than some people might think. You basically have a limited set of constraints to work to and then the rest is a matter of personal style. The key constraints to the design process are the box volumes for the mid-range and the woofer, the height of the tweeter mid-range mid-line above the floor and avoiding simple integer ratios in box dimensions.
The height of the mid-line above the floor is important as you would optimally like that mid line to be at ear level in a normal seated position. Depending on the height of the individual and the type of seating that can vary from between about 75cm to 115cm above the floor. In my design I chose a height of 98cm above the floor.
The importance of avoiding integer ratios in box dimensions is out of consideration of standing waves. Just as in room acoustics, a loudspeaker cabinet that has integer ratios of dimensions will exhibit strong internal standing wave behaviour which could have a major negative influence on the performance of the woofer and the mid-range unit for that matter.
The basic plans I came up with excluding details of the mid-range flange is shown below.
The tweeter is mounted 90cm above the floor and the mid-range and tweeter centres are 16cm apart giving the mid line height of 98cm. The internal dimensions of the woofer enclosure (ignoring the mid range box for now and assuming a 2cm MDF thickness) are 23cm width, 30.5cm deep and 112cm tall. That gives a depth on width ratio of 1.326, a height on width ratio of 4.870 and a height on depth ratio of 3.672. None are simple ratios as required.
The drivers themselves are then mounted as close as practically possible to one another. In this regard, cursory inspection of the plans might suggest that the tweeter could be closer to the mid-range, however, you need to consider that the mid-range unit will be set forward of the tweeter by a flange and we need to allow some extra space to place the flange cusp that we need to avoid excessive colouration of the tweeter response. The woofer position is not so sensitive and doesn’t affect the response that much but we should leave some space between the driver port and box walls to avoid introducing asymmetrical air mass loading on the cone, which in turn my lead to increased distortion. We should also avoid mount the woofer too close to the floor as this will result in a substantial lift in bass coupling with the room which in turn make our speakers sound muddy. The vent should also not be too close to the floor nor too close to the woofer to again avoid too much asymmetry in air mass loading on the woofer.
The internal dimensions of the mid-range enclosure are 23cm width, 29cm height, 8cm depth giving a width on depth ratio of 2.875, a height on depth ratio of 3.625 and a height on width ratio of 1.261. Again, none are simple ratios. The box volume is 23×29×8=5336 cubic centimetres or 5.336 litres. Our design was for 5 litres but a portion of the box volume with be taken up by the volume the drivers displace and the volume displaced by the box stuffing. 0.36 seems like a reasonable guestimate of these displacements and in any case, the driver response is not very sensitive to changes in box volume so any small errors should have minimal impact on the final result.
The woofer enclosure volume is (23×30.5×112)-(23×31×10)=71438 cubic centimetres or 71.438 litres. Our design goal was a figure of 70 litres and the 1.438 extra litres is a reasonable allowance for the volume displaced by the driver, the vent, the crossover and the box stuffing. Again, the design is not very sensitive to changes in box volume so any error should have a minimal effect. The major effect would be on box tuning but this will be tuned correctly through measuring the impedance of the woofer mounted in the cabinet and adjusting the vent length until the impedance shows the correct box frequency.
That is the basic cabinet design. You could make it more elaborate by using angled walls to further reduce standing wave issues and a more sculptured look for aesthetic reasons. It all comes down to how much effort / cost you can justify. More elaborate designs add a little bit of extra performance but generally require considerably more effort to construct so I’ve stuck with something simple.
Suffice to say that I’ve had enough experience with making speakers in the past to know that it is worth testing a physical realisation of the design first before settling on it. Unforeseen consequences are all to common, even in the case of the extensive modelling that I’ve done.
The biggest question for me in the whole design which isn’t accurately represented in the modelling is the tweeter set-back. The modelling simply assumes that you can realise the time delay required without any impact on the driver frequency responses, which is a quite big assumption. The manufacturer specs assumes the drivers are mounted on a flat baffle. A flat baffle isn’t possible because we need to have the tweeter set back (or the mid-range set forward) by 25.4mm. This necessarily introduces boundaries in the baffle that can reflect sound and alter the driver responses.
My original naive idea was to set back the tweeter in a recess smoothly shaped to try an minimise diffraction effects on the tweeter response. I’m made the prototype shown below to test it as I wasn’t at all confident it would work satisfactorily.
As it turns out my naive idea was a bad idea because it substantially alters the frequency response by focusing sound in the 5khz region forward. This is completely unsatisfactory from my point of view so I had to re-think my approach.
My second line of thinking revolved around the idea of setting the mid-range forward the desired amount and scattering any sound directed at the flange setting the driver forward. Ideally something like a quadratic diffuser would be ideal but the space required for such an approach simply isn’t there. The next best option would be to push the wave front out into space in a direction where it would least effect the driver frequency responses. To that end I ended up with the flange shown below.
Here I’ve introduced a pointed cusp on the flange to direct incident sound towards the sides and have avoided cutting the flange with a tapper so as to avoid directing reflections forward. Measurement and listen both confirm that this approach has minimal effect on the tweeter driver response so it will be the approach I take in the final design. Knowing the changes required I have to go back and modify my final cabinet design which assumed the former approach.
The crossover board design and construction is a complicated affair. Not wanting to do any circuit board etching I needed to figure out how to lay out the components on a board with copper cuts made to form the tracks, all on a board that is small enough to fit through the hole for the woofer. Given the highly over rated polycarbonate capacitors (400V working) the capacitors are huge and take up quite a bit of real estate as do the inductors. That and having circuit board stock of 305mm square to work from meant that putting the entire crossover on the one board was simply not possible. Instead I made a prototype board that has the mid-range and tweeter parts on a board measuring 305 by 152mm (narrow enough to fit through the woofer port) with little room to spare. The assembled board is shown below,
Along with the prototype board you can see an L-Pad that I’ve inserted to figure out the correct tweeter padding needed for a flat on axis response. I don’t know why it is, but every time I’ve attempted to get the padding right from manufacturers specs alone it’s invariably been wrong and this time is no different in that respect. The tweeter needs an additional 3 dB of padding to bring it in to line with the mid-range output so I’ll have to also slightly modify my crossover design to suit. In other respects the crossover region for the mi-range and tweeter units are seamless, leading to a neutral sound, however it’s a bit difficult to make an realistic evaluation of the sound quality in the absence of the woofer output. That will have to wait but it is looking good so far.
As far as the behaviour of the crossover (when the padding is right) it is pretty much in line with what the modelling suggested would be the case : the vertical plane off axis response on the side of the tweeter without the mid-range being notched and the response on the other side more uniform. That means that at least the order of the drivers in my cabinet design are correct. The mid-range appearing at the top, followed by the tweeter below and the woofer below that. Actually, it would be more ideal for the woofer to be mounted next to the mid-range unit (to minimise the space between them) but since the crossover frequency is so low (150Hz) and the wavelengths at that frequency so big (2.3m), it makes little difference which way I arrange them so for the sake of keeping the cabinet from getting too tall I’ll be mounting the woofer below the tweeter.
That’s it for now. I need to go and do some design modifications and work on the woofer cross-over circuit board and build the cabinets…