Driver measurement help

[Wheels]

Could someone post a file of your driver measurements? I don't care what driver. I was doing some crossover work in speaker workshop and the raw data just seems real fuzzy. It smooths out pretty much ok, I just have one really bad spike at ~3.5k and it is exaggerating a dip and I think making the graph look worse than it really is.

I attached the file that has been giving me problems along with the 30 degree response if anyone has an idea of what could be going wrong. I had to change the extension from .frd to .txt so it would upload.

its a Selenium D220ti driver mated to a selenium HC23-24 horn. Dayton EMM-6 mic, Fuzzymeasure trial measuring software. speaker located in the middle of the room in cabinet, measurements taken at 24"

And if you were wondering, a friend asked me to build a PA speaker for him for a jr high class at church. He mainly wanted it to get loud for cheap and he didn't care for excessivly high fi. his current speakers in there are 20+ years old and worn out. I played with some filtering in a .wav file to get an idea of what they would sound like and was pleasantly suprised.

Thanks in advance

[5th element]

You're not doing anything wrong with regards to measuring the selenium, the dip at 3.5k is there in the datasheet for the horn, but oviously smoothed to make it less severe. Horns can show suckouts when directly on axis, with the dips disappearing when going off axis.

http://www.parts-express.com/pdf/264-318.pdf

With regards to the measuring and the raggedness, this is because you're not using gated measurements. The below images show on the left hand side the raw impulse response that is taken when performing a standard measurement. On the far left you can see the initial stimulus and then decay, but as time goes on you see small spikes appearing. I have pointed at the spikes with the white lines, these are room reflections and are what cause the FR to become jagged/ragged.

The first image shows the impulse response on the left and on the right you've got the corresponding frequency response. This is using 24/octave smoothing too, without it things are much worse.

The second image down on the left is gated. As you can see I've set a start and end point on the impulse response with the stuff outside of these points greyed out. Here I've removed/gated out most of the reflections and as you can see on the FR on the right, it is a lot smoother.

The third image down on the left goes to an extreme and shows what happens if you remove practically all of the reflections. It's corresponding FR on the right is completely un-smoothed, but as you can see it is very smooth itself with only a couple of bumps here and there.

One thing to point out is that the accuracy at low frequencies diminishes as the gate is make smaller, you can easily see this in the graphs below and to make life easier for you ARTA also indicates the lower limit on the FR graphs with the horizontal yellow line that runs along the bottom of the graph. To get around the low frequency limitations of gated measurements, one usually takes a near field (mic next to woofer cone) measurement, as the mic is right next to the cone there are no reflections to speak of so you can open the gate completely and end up with good low frequency accuracy. This is great, but it also presents a problem, it does not show the high frequencies with particularly good accuracy. What one normally does is perform gated far field measurements to get accurate high frequency data to around 500-800hz, then you perform a near field measurement and splice the nearfield measurement onto the farfield one. This provides the best of both worlds, but sometimes requires the addition of simulated baffle step.

[Wheels]

Ahhh. Thank you so much. I guess I need to zoom in more to see the room reflection. It didn't dawn on me it would be that small.

That brings me to more questions, A) should I remeasure before committing to a crossover? My smoothed graphs were fairly similar to the manufactures specs, I am just worred the added "fuzz" will mess things up. B) I've heard of splicing the response in, but how do you make sure the level is matched C) wouldn't a nearfield response eliminate the port effect of the low end response?

also, I attached the graph I obtained with using the 30 degree response curves. the woofer is the Eminence 12A Delta Pro. I'm fairly happy with it, I don't think the extra brightness on axis will be too much. They are going to be mounted against a wall so I expect a little extra low end to end up with the smile response a lot of people like. But that is just my assumption.

[5th element]

A) should I remeasure before committing to a crossover? My smoothed graphs were fairly similar to the manufactures specs, I am just worred the added "fuzz" will mess things up.

It isn't just fuzz that the reflections add in, they add in peaks and dips that aren't there in reality. Smoothing can help against this, but one strong reflection could put a sizeable bump in one place that would grossly affect the tonal balance. If you're aiming for 'perfection' one could say, then the old 'rubbish in rubbish out' comes to mind, you'll be shooting in the dark. Whether or not it's worth measuring again in this design is another matter. You've mentioned that the loudspeaker probably wont be used for critical listening, but as the loudspeakers they are replacing have been used for the last 20 years you could assume that these might be used for just as long. 20 years is along time, if a loudspeaker I was designing was going to be used for that long, I'd want to go the extra mile to make sure they are as good as they can be for the given budget.

B) I've heard of splicing the response in, but how do you make sure the level is matched?

You use the far-field response as the reference. When you splice in the near-field you alter it to match the far-field. Typically one applies a simulated baffle step response to the near-field. The far field should already contain the lumps and bumps that cabinet diffraction will cause so that's already taken care of. All that's required is the near field response match the farfield, which means it needs a simulated baffle step response adding to it to match the low end roll off of the far-field.

wouldn't a nearfield response eliminate the port effect of the low end response?

Yes and you can measure the port on its own and add it to the near field overall response. This however is completely unnecessary in 99% of situations because you're not going to be doing any crossover work down where the ports contribution in significant.

also, I attached the graph I obtained with using the 30 degree response curves. the woofer is the Eminence 12A Delta Pro. I'm fairly happy with it, I don't think the extra brightness on axis will be too much. They are going to be mounted against a wall so I expect a little extra low end to end up with the smile response a lot of people like. But that is just my assumption.

Ah if they are going to be mounted on a wall then they will require minimal baffle step compensation, if any. This simplifies things considerably as you wont need to make any near field measurements. Even with a small gate you can accurately get reflection free data down to 500hz, which is more then adequate for accurate crossover design with a target crossover @ around 2khz.

What I would be concerned about, using the 30 degree data, is how much the 12" mid/bass will have been rolled off by 2khz. This will add a rising bump of around 6dB around the 2khz xover frequency on axis and will make the on axis sound quite fatiguing. We are far more tolerant of dips in the FR then we are of peaks. I would suggest designing the xover around the on axis plots for the Delta and possibly the off axis plots for the selenium horn. The off axis low end response of the horn (around the xover frequency etc) shouldn't be too different from the on axis.

[cjd]

Remember that, as soon as you start splicing data, you also need to be modeling with minimum phase (and accounting for z-axis offset)

I always run on-axis through 45 degrees off-axis and model them all with the same crossover to get a better handle on how things will work. Usually 0, 5, 10, 15, 30, and 45. For these, I never do any splicing - always done without moving speaker or mic so phase as-measured is exactly relative, and I largely ignore anything below where gate frequency limits accuracy. This only becomes difficult if you're doing a 3-way with a crossover low enough to be in this range.

C

[Wheels]

yeah, it took me a little bit to figure out how to get minimum phase in speaker workshop. the weird part is I am using my HT reciever and sending audio out over hdmi so my delay is huge (40-50 ms) and it varies every time. That shouldn't matter as long as I adjust the .frd file with the correct delay in speaker workshop for each graph?

And also, my baffle is 16.5 inches across, including sides. I was wanting to account for partial baffle step since I put the top hat stands in the bottom and they could be used on a tripod away from the wall. That would put the BSC around 250 hz if I figured it right. retaking measurements in my living room I had to go with ~16ms half hamming window to keep reflections to a minimum. That really doesn't get me low enough to check baffle step issues?? I want to say the resolution was 75hz ish. It would give me an idea, but it doesn't seem like it would be too accurate. so how do you simulate baffle step?

Thanks guys for the help. I remeasured and gated the response it it looks much much better. this time the phase is legible across the response.

[5th element]

When measuring a loudspeaker in the far field you get what could be described as two effects occurring. The first is the standard 6dB loss as you transition from 2pi into 4pi radiation. In your case the centre frequency is about 150hz with the transition and diffraction effects starting @ around 1000hz. The second effects are the diffraction ripples themselves and these occur around the onset of baffle-step. In your case you get the standard peak/dip combo, the peak being rather large and centred at around ~600hz with the dip occurring at a little over 1k.

The point to all of this is that the peak/dip combo is what you're interested in capturing from the far-field measurements. You should be able to get accurate far-field data to around 200hz without too much issue. This will therefore show you the diffraction ripple peak/dip combo, but it will also show you the start of the baffle-step transition.

What you do now is take a near-field measurement that shows the low end response and apply a standard baffle step model to it. This will therefore show the full 6dB loss and if you overlay it with the far-field measurement the two (if they correctly level matched) should match up fairly well. That is the baffle step model applied to the near field measurement should start to follow the far fields measured baffle-step response at around 200-300hz. You should end up looking at a 6db transition with the near field measurement representing the effect at low frequencies and the far-field measurement representing the effect at high frequencies. You then pick a frequency where the two match up really well and then splice them together.

As CJD mentioned the phase is extremely important and there is an easy way to make sure that you've got this right. What you are after here is phase data that accurately represents the relationship between the tweeter and the mid/bass driver so that the crossover will be simulated accurately. This is not difficult to do, but requires a precise set of measurements.

1) Put the microphone on what would be the standard listening axis and at distance that would represent what a decent listening distance might be.

2) If the listening axis is on the tweeters axis take a measurement. Now set the gate to start just before the impulse response and end just before the first reflection. If you're listening axis is on the mid/bass take this first measurement and the initial setting of the gate with the mid/bass driver. Save this response.

3) Leave the microphone in exactly the same place and measure the other driver (if you measured the tweeter in the previous step, now measure the mid/bass), do not touch the gate, it needs to be left exactly where it was for the first measurement. Now save this. (If the point of the first reflection happened to arrive earlier you can alter the end point of the gate to exclude it, but do NOT change the start point).

4) Connect both of the drivers in parallel and measure them together. Again do not touch the gate and save this.

What you've got now are individual measurements for the tweeter and the mid/bass and as the conditions that each measurement were taken in was identical, the phase relationship between the two should have been accurately measured too. The measurement with the two drivers together shows how the two drivers sum when playing together and also shows how the two individual measurements should sum when the simulator attempts to sum them.

What you do now is load the tweeter and mid/bass response into your simulator and then you then enter the relevant offset between the drivers in the X and Y axis. If you measured the drivers on the tweeter axis then X and Y for the tweeter should = zero. You then alter the X and Y position for the mid/bass relative to the tweeter, if the centre of the mid/bass happens to lie 30cm below the centre of the tweeter then you set the Y parameter of the mid/bass to -30cm.

If you can set the measurement distance within the simulator then do so. The simulator should now be showing you a combined response, where it calculates/simulates what the two drivers should produce when summed together.

Now you want to load in the measurement of the two drivers combined and you want to set it as some sort of target response. What you should be looking at now is the simulators simulated sum of the two drivers and the real world measured sum of the two drivers. What you do now is alter the Z parameter for one of the drivers (I usually pick the tweeter) until the simulated sum and the real world sum match one another. Once you've got a good match, you're ready to start designing a crossover.

If you are interested in combining a near-field low frequency measurement with a far-field high frequency measurement then you use the farfield measurement, as taken above as the reference. That is you use the measurement taken in step two or three as the high frequency part. You then alter the level and the delayphase of the near-field measurement until it matches the far-field one. This way, as you've kept the far field measurement the same, you've still preserved the relationship between the two measured drivers, only you've tacked on a near-field measurement to the lower half of one of them for better low frequency accuracy.

[Wheels]

OK. I I followed everything except the identical gate part. I'm not sure how to not alter the starting point of the gate. on my last set of measurements my impulse delay ranged from 31.73 ms to 40.33 ms. so I'm going to have to have 10ms of dead air before one of my measurements in that case? Or should I run the sweep until I get 2 start times fairly close together?

And thanks again for all your time here. I really appreciate the help.

[5th element]

Arg that damned processor. You really need a measuring system that hardware-wise doesn't change its delay from one measurement to the next. If it is doing this then getting accurate phase data is going to be a trifle problematic. If what I think you're saying is occurring then you'd be better off using anything else as a a source, any computer sound card should do no matter how crummy, just use its analogue outputs if you can.

What I think is happening is you are taking a measurement of say the tweeter and you set the gate, but then you take another measurement of the tweeter and the gate is now in the wrong place, because the impulse has been delayed by a different amount due to hardware.

I have experimented a little with loudspeaker workshop in the past but I've never really been a fan of it. You could try using ARTA to take measurements, the free version will work just fine and it can export the measurements you make too. It's rather easy and intuitive to use although you will have to familiarise yourself with it. It will easily let you see though if the delay is software related.

If you want to give ARTA a quick try I can give you a short walk through of how to make a loudspeaker measurement, although I dare say you'd probably be able to figure this out for yourself in a few minutes

Edit - I thought I'd just clarify, that what you're after when you're making these measurements is the precise difference between the start of the impulse responses when you're measuring the tweeter and the woofer separately. 10ms is = to around 3 meters of travel where sound is concerned. The usual ball park figure for the time difference/distance between the tweeter and woofer should be more like 3cm, so your hardware is adding in around 100 times more in terms of experimental uncertainty then the drivers should ideally be showing. Of course having the horn loaded tweeter will push the acoustic centre back so you could see something a little more then 3cm, but I doubt it'd be anything more then around 10cm. This variable time delay in the hardware would also help explain why there was such a monumental difference between the phase of the 30 degree off axis tweeter measurement and the on axis measurement you posted before. I mean sure I didn't expect them to be coherent, but nevertheless they should be within a few micro seconds normally, to one another.

[Wheels]

Yep, you nailed it. My impulse delay was all over the board.

OK. hmm. I was using FuzzMeasure for Mac. I guess before I try anything too drastic I could connect the 1/8" out to the RCA on my amp and see if that fixes it. I know my amp has a straight analoge mode and that should fix it.

I might end up trying ARTA, but that will require moving a computer or disassembeling my HT even farther, and the only computer I have running windows has fan noise that I'm worried will mess up the data too.

Now also, Speaker workshop has the ability to manually enter the delay for each graph, but then the position of the impulse relative to the start of the gate is still only as accurate as I can set it. I've tried that and didn't really seem to get good results, was that my error or just a bad idea? I still think I'm going to go 1/8" out though.

[5th element]

Yep, you nailed it. My impulse delay was all over the board.

Well that sucks. If it were out by a few micro seconds you'd be able to compensate when setting the Z axis in the simulator. It wouldn't be perfect, but it'd be decent enough. A few ms though :E

OK. hmm. I was using FuzzMeasure for Mac. I guess before I try anything too drastic I could connect the 1/8" out to the RCA on my amp and see if that fixes it. I know my amp has a straight analoge mode and that should fix it.

FuzzMeasure should be fine, I've tried using speaker workshop in the past and found it the most unfriendly piece of software to use, so I was wondering if it had any limitations etc. Using the standard sound card output could be a much better choice, give it a go and see what you get.

I might end up trying ARTA, but that will require moving a computer or disassembeling my HT even farther, and the only computer I have running windows has fan noise that I'm worried will mess up the data too.

If the loudspeakers are measured at a decent volume then a bit of fan noise shouldn't be much of a problem. From what I've heard though Fuzz measure is a really nice program, I think JonMarsh has used it and said it was pretty good? Unless you really want to you could try using ARTA, I love it, very user friendly and can do a whole heap of things.

Now also, Speaker workshop has the ability to manually enter the delay for each graph, but then the position of the impulse relative to the start of the gate is still only as accurate as I can set it. I've tried that and didn't really seem to get good results, was that my error or just a bad idea? I still think I'm going to go 1/8" out though.

You can't accurately set the delay unless you know how much delay the hardware added in the first place and as it's variable in your case it'd be like shooting in the dark. I know that a two channel measurement can correct for any electrical effects such as any roll off intrinsic to the output side of the measurement system. This way you can measure a tweeter with a cap in series and its effect will be ignored. I don't know however if this is capable of cancelling out variable delay times, I mean the computer would know when it sent out the impulse response so it could measure the time delay between its computer side creation and the actual creation on the output of the amplifier, then deduct this time off of the start of the impulse response, that way even if the time delay changed it would be effectively removed.

[Wheels]

well, I finally got something that resembled the two drivers in parallel when summing the freq responses together. It wasn't perfect, but I think it will work.

Thanks for all the help guys. I definitely learned a lot during all this. I have decided to name the project the Big O's, 1) he wants an orange front baffle with black everywhere else. It looks pretty cool. and 2) I lost track of how many times I did something and went "Oh" and got to re-figure it all out because I didn't have it right. Once I get some time I'll post the pics of it all

[5th element]

well, I finally got something that resembled the two drivers in parallel when summing the freq responses together. It wasn't perfect, but I think it will work.

Excellent, sometimes the match isn't perfect, but as long as it's decent you usually do fine.

I have decided to name the project the Big O's, 1) he wants an orange front baffle with black everywhere else. It looks pretty cool.

Where are the loudspeakers going to be used? Black and orange :E

Once I get some time I'll post the pics of it all

I can't wait to see them :T

[Wheels]

They are being used in a very brightly colored room. the wall they are going to be sitting in front of is bright green so he wanted the front orange. the rest is black so they look mostly normal. its a lot of accent type lighting so very rarely will they be in a fully lit room so the bright colors help stuff stand out. The room is for the Jr. high age kids at church so bright colors keep it fun (I guess... I'm an engineer I don't know anything about that)