Hi,
I'm playing around with Speaker Workshop, trying to model crossovers and learn how the software works. I have a few questions that I would appreciate help with. I'm trying to design a passive crossover between an Audax PR170M0 and a Fostex FT17H. These will be mounted to an open baffle (that's the plan so far).
First, I used offset.xls (don't remember where I found that software) to try and find crossover frequencies/slopes to give me a smooth frequency response. I assumed that the acoustic center of the horn tweeter will be at the diaphragm/voice coil, and for the midrange driver I took the midpoint of the magnet. Feeding this into the spreadsheet and playing with values, this is what I got:
I cropped that too small - those are 2nd order Butterworth filters, 2900Hz lowpass and 5400Hz highpass, and a 3cm offset. That was the smoothest response I could get while staying within the limits of the drivers and avoiding shallow but broad dips.
I used SPL Trace to create FRD and ZMA files for my drivers. I used the network optimizer tool in Speaker Workshop, as well as PCDC.xls (Passive Crossover Design Calculator, most of this software can be found on FRD Consortium website) to pick L and C values to get to this target response. I put an L-pad on the tweeter to bring its level down closer to the midrange. The final network consists of one L and one C on each driver, and two Rs on the tweeter for an L-pad.
The best result I could manage:
It looks pretty horrible until you see that the scale on the left is 2dB. If I add in the individual driver responses, that makes the scale become something more sensible, and the result looks like this:
I then played with the component values manually, trying to flatten the transition. This is one of my results:
Looks better at this resolution, but when I zoom in by removing the individual driver curves:
Which looks worse than the previous crossover, though this is at 1dB resolution so maybe it isn't all that worse.
So, my basic question - which resolution level makes more sense, and correlates better with what I'll hear in-room? Should I be trying to flatten 1-2dB peaks and dips, or is it more important to get the broad shapes correct?
Also, if there are any big holes in my approach, please point them out. I know I haven't looked at off-axis response and phase and impedance. I'll try those later once I figure out the first step, on-axis response (even though I know they all go hand-in-hand). I'll also measure my drivers once I've built the baffle and mounted them, and re-do all of this with that frequency response data. Which means I need to learn how to correctly measure them too.
Thanks for looking,
Saurav
Edit: The tweeter crossover ended up with just a series capacitor (the optimizer suggested 1000H for the inductor, so I took it out).
I'm playing around with Speaker Workshop, trying to model crossovers and learn how the software works. I have a few questions that I would appreciate help with. I'm trying to design a passive crossover between an Audax PR170M0 and a Fostex FT17H. These will be mounted to an open baffle (that's the plan so far).
First, I used offset.xls (don't remember where I found that software) to try and find crossover frequencies/slopes to give me a smooth frequency response. I assumed that the acoustic center of the horn tweeter will be at the diaphragm/voice coil, and for the midrange driver I took the midpoint of the magnet. Feeding this into the spreadsheet and playing with values, this is what I got:
An externally hosted image should be here but it was not working when we last tested it.
I cropped that too small - those are 2nd order Butterworth filters, 2900Hz lowpass and 5400Hz highpass, and a 3cm offset. That was the smoothest response I could get while staying within the limits of the drivers and avoiding shallow but broad dips.
I used SPL Trace to create FRD and ZMA files for my drivers. I used the network optimizer tool in Speaker Workshop, as well as PCDC.xls (Passive Crossover Design Calculator, most of this software can be found on FRD Consortium website) to pick L and C values to get to this target response. I put an L-pad on the tweeter to bring its level down closer to the midrange. The final network consists of one L and one C on each driver, and two Rs on the tweeter for an L-pad.
The best result I could manage:
An externally hosted image should be here but it was not working when we last tested it.
It looks pretty horrible until you see that the scale on the left is 2dB. If I add in the individual driver responses, that makes the scale become something more sensible, and the result looks like this:
An externally hosted image should be here but it was not working when we last tested it.
I then played with the component values manually, trying to flatten the transition. This is one of my results:
An externally hosted image should be here but it was not working when we last tested it.
Looks better at this resolution, but when I zoom in by removing the individual driver curves:
An externally hosted image should be here but it was not working when we last tested it.
Which looks worse than the previous crossover, though this is at 1dB resolution so maybe it isn't all that worse.
So, my basic question - which resolution level makes more sense, and correlates better with what I'll hear in-room? Should I be trying to flatten 1-2dB peaks and dips, or is it more important to get the broad shapes correct?
Also, if there are any big holes in my approach, please point them out. I know I haven't looked at off-axis response and phase and impedance. I'll try those later once I figure out the first step, on-axis response (even though I know they all go hand-in-hand). I'll also measure my drivers once I've built the baffle and mounted them, and re-do all of this with that frequency response data. Which means I need to learn how to correctly measure them too.
Thanks for looking,
Saurav
Edit: The tweeter crossover ended up with just a series capacitor (the optimizer suggested 1000H for the inductor, so I took it out).
sreten.
We are amplitude oriented, so if the peaks/dips are high Q, then we tend to only 'hear' the peaks, so if they present a ~flat FR you're OK. As for resolution, +/- 2dB is exceptional. I used +/-3dB and all were pleased.
WRT horn acoustic centers, as you go down in frequency, it moves towards the mouth, so at 5900Hz it will be somewhat in front of the VC.
GM
WRT horn acoustic centers, as you go down in frequency, it moves towards the mouth, so at 5900Hz it will be somewhat in front of the VC.
GM
Oh I don't think I have +/-2dB or even +/-3dB across the board. I was trying to figure out when a peak/dip would be considered bad enough that I'd have to treat it individually. If I understood your point about peak/dip Q right, then for the sharp/narrow peaks I should try and line those up with the flatter smoother regions at other frequencies. That makes sense.
Thanks,
Saurav
Thanks,
Saurav
Thanks. This is what I have so far. This is purely modelling based on printed frequency response and impedance curves, I haven't mounted and measured any drivers yet. I'm sure some (or a lot) of this will change when the drivers actually go onto the baffle. Anyway:
I started out with component values calculated by the software, and then I changed them (at least the Ls and Cs) to values that I know I can get, or I already have. I'll try and bring that one peak down some more. It's in the on-axis FR plot of the Audax driver, though some people have said that those peaks aren't all that audible, and someone said he's seen an FR plot where those peaks aren't even present. We'll see.
Does this look OK so far? I went for target 2nd order Butterworth acoustic slopes, at 2900 and 5400 (got those values from playing around with offset.xls). The schematic includes my estimated driver offset. Should I be aiming for steeper acoustic slopes? I thought those frequencies looked good, because the Audax shouldn't start beaming too much at 2900. The resistor across the midrange is the zobel, Speaker Workshop calculated a 0F capacitor in series so I took that out. I don't think the phase plot can be trusted, because my FRD and ZMA files were generated using SPL trace so they say 0 degrees at every frequency.
The low pass will be an active XO at 500Hz, probably 4th order L-R. The baffle is 14" wide, which puts the BSC corner at 325Hz, I can shift it a little lower by adding small wings to the sides, so I think that end of it should be OK.
Can I get Speaker Workshop to show me what the off-axis response will look like? Or do I just use the 30 and 60 degree FRD files with the same crossover network and that's how I get my off-axis response?
Thanks in advance,
Saurav
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
I started out with component values calculated by the software, and then I changed them (at least the Ls and Cs) to values that I know I can get, or I already have. I'll try and bring that one peak down some more. It's in the on-axis FR plot of the Audax driver, though some people have said that those peaks aren't all that audible, and someone said he's seen an FR plot where those peaks aren't even present. We'll see.
Does this look OK so far? I went for target 2nd order Butterworth acoustic slopes, at 2900 and 5400 (got those values from playing around with offset.xls). The schematic includes my estimated driver offset. Should I be aiming for steeper acoustic slopes? I thought those frequencies looked good, because the Audax shouldn't start beaming too much at 2900. The resistor across the midrange is the zobel, Speaker Workshop calculated a 0F capacitor in series so I took that out. I don't think the phase plot can be trusted, because my FRD and ZMA files were generated using SPL trace so they say 0 degrees at every frequency.
The low pass will be an active XO at 500Hz, probably 4th order L-R. The baffle is 14" wide, which puts the BSC corner at 325Hz, I can shift it a little lower by adding small wings to the sides, so I think that end of it should be OK.
Can I get Speaker Workshop to show me what the off-axis response will look like? Or do I just use the 30 and 60 degree FRD files with the same crossover network and that's how I get my off-axis response?
Thanks in advance,
Saurav
And if my assumption about modelling off-axis FR is correct, this is what that should look like after 1/3 octave smoothing at 0, 30 and 60 degrees (the graph in the previous post is unsmoothed):
An externally hosted image should be here but it was not working when we last tested it.
Better? This is 1/3 octave smoothed.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Fostex
Hi Saurav,
Don't want to put a damper on the proceedings, as it looks like you are making good progress, but the Fostex is really a supertweeter.
The resonance is high (Fostex shows 2.7K cut-off) and it will sound pretty bad crossed below 4K. Even there you will need at least an 18 dB 3rd order slope to keep the unit from sounding bad or frying. Fostex recommends use at 5K and above for this reason.
If you do try a 4K xover, you will be unable to feed more than a few watts safely.
I'd wait until I had heard the Audax well broken in and run up high to find the best crossover point.
You may encounter some beaming (narrowing of dispersion)above 4K, but it may not be objectionable, nor is the slight break-up shown on the frequency response graphs at this frequency.
If your intent is use this in a home setting with tube gear of low power, you can push the limits a bit, but going low with the s-tweeter is not wise. I might consider a ribbon from Fountek or Aurum Cantus crossed in at 2K - 2.5K or so. They are very fragile, but sound much better than the Fostex. I have the Radio Shack equivalent, and though it may not be a direct duplicate it has the same specs. Not very good sounding this low.
A small compression driver and horn would also work with necessary padding of the output. Fostex makes a good adjustable L pad for such purposes.
Tim
Hi Saurav,
Don't want to put a damper on the proceedings, as it looks like you are making good progress, but the Fostex is really a supertweeter.
The resonance is high (Fostex shows 2.7K cut-off) and it will sound pretty bad crossed below 4K. Even there you will need at least an 18 dB 3rd order slope to keep the unit from sounding bad or frying. Fostex recommends use at 5K and above for this reason.
If you do try a 4K xover, you will be unable to feed more than a few watts safely.
I'd wait until I had heard the Audax well broken in and run up high to find the best crossover point.
You may encounter some beaming (narrowing of dispersion)above 4K, but it may not be objectionable, nor is the slight break-up shown on the frequency response graphs at this frequency.
If your intent is use this in a home setting with tube gear of low power, you can push the limits a bit, but going low with the s-tweeter is not wise. I might consider a ribbon from Fountek or Aurum Cantus crossed in at 2K - 2.5K or so. They are very fragile, but sound much better than the Fostex. I have the Radio Shack equivalent, and though it may not be a direct duplicate it has the same specs. Not very good sounding this low.
A small compression driver and horn would also work with necessary padding of the output. Fostex makes a good adjustable L pad for such purposes.
Tim
Well, the cap on the tweeter is set to start rolling it off around 5K, isn't it? I modelled it for a 2nd order rolloff at 5.4K, but I've messed with the values since then so I don't know what the -3dB point is now. I think I see what you're saying, the tweeter isn't down all that much at 5K. Alright, I'll play with this some more this evening. And you're right, I should wait until I've run in all the drivers before making a final decision. If the Audax sounds fine being used up to a higher cutoff, I'll shift things up in that direction.
Slope
I just don't trust a single cap to do the job. Although the slope looks good, the crossover is really at 3.6 or 3.7K Hz. See what moving above 4K does.
I used these (RS units) with just a .47 uf cap(!), and was still surprised at the amount of output low. Also, the modeling will not show the amplitude correctly as these are 2 - 3dB louder than what is shown on the specs. You may need to add more series resistance.
Naturally, measurements of the actual units are the key.
Tim
I just don't trust a single cap to do the job. Although the slope looks good, the crossover is really at 3.6 or 3.7K Hz. See what moving above 4K does.
I used these (RS units) with just a .47 uf cap(!), and was still surprised at the amount of output low. Also, the modeling will not show the amplitude correctly as these are 2 - 3dB louder than what is shown on the specs. You may need to add more series resistance.
Naturally, measurements of the actual units are the key.
Tim
OK, if they're 2-3dB louder, that will change things. I modelled an L-pad for this tweeter and it gave me a 75 ohm paralel resistance, so I took that out and just kept the series resistance in there.
I'm still not 100% sure I see where you're getting your readings from. To me, it looks like the peak tweeter output is around 98dB, and it's down to about 95dB by 5K, and about 85dB by 2.5K.
Maybe this is the confusion - the red and blue are the tweeter and midrange with their respective XO networks, not the raw response of the drivers. I could post another image this evening with the bare driver response overlaid if that'll be useful.
But, I see your point. I'll play with a higher order electrical network on the tweeter and see how that looks.
I'm still not 100% sure I see where you're getting your readings from. To me, it looks like the peak tweeter output is around 98dB, and it's down to about 95dB by 5K, and about 85dB by 2.5K.
Maybe this is the confusion - the red and blue are the tweeter and midrange with their respective XO networks, not the raw response of the drivers. I could post another image this evening with the bare driver response overlaid if that'll be useful.
But, I see your point. I'll play with a higher order electrical network on the tweeter and see how that looks.
Slope
I agree with the 12-13 dB down point of the tweeter and your results so far. I meant 10dB further down.
I think the results your modeling are accurate but not a true reflection of reality by the amplitude being higher, and the assumption that they will sound the same at 4K Hz as they do at 5K, amplitude equal. They don't. They sound worse to me.
I'd use 3rd order at 5K Hz, then verify when you get the units, or go with a different tweeter. It may be a bias on my part, so wait until you receive them and run your tests.
Check the return policy from your source, though, in the event they don't work out. For the dollars, I can't suggest a substitute that I've tried, but the higher priced ribbons (Aurum Cantus G2 and 2si) sound better to me.
Tim
I agree with the 12-13 dB down point of the tweeter and your results so far. I meant 10dB further down.
I think the results your modeling are accurate but not a true reflection of reality by the amplitude being higher, and the assumption that they will sound the same at 4K Hz as they do at 5K, amplitude equal. They don't. They sound worse to me.
I'd use 3rd order at 5K Hz, then verify when you get the units, or go with a different tweeter. It may be a bias on my part, so wait until you receive them and run your tests.
Check the return policy from your source, though, in the event they don't work out. For the dollars, I can't suggest a substitute that I've tried, but the higher priced ribbons (Aurum Cantus G2 and 2si) sound better to me.
Tim
I looked into the AC ribbons, and that's more money than I want to spend on my first ever speaker design attempt. I'll keep your words in mind, and most other threads I've seen about the FT17H do cross them over higher. I'll try modelling a 3rd or 4th order acoustic slope on the tweeter and see where that gets me.
Thanks a lot for the help.
Thanks a lot for the help.
OK, how about this? 3rd order electrical on tweeter, 2nd order electrical on woofer, tweeter polarity inverted. I couldn't find a way to avoid that - is it important to find an XO that works without requiring this polarity inversion? This helps the frequency response, obviously, but I'm not sure if it hurts anything else (step? square wave?) and if they're important.
Schematic, unsmoothed response, 1/3 octave smoothed response with individual driver + network responses added in:
Not as smooth as before, and I also lost about a dB of overall efficiency, but the tweeter is much more attenuated at the lower frequencies. Also, if those breakup mode peaks aren't quite as bad in real life as they are in my SPL trace, then that will help this.
For some reason, calculating the zobel using PCDC.xls (uses Re and Le) and inserting that manually gave me better results than using Speaker Workshop's internal calculators (which always put the C at 0F). Maybe I should try PCDC for the component values too, instead of using Speaker Workshop to do the math. For now, I'll assume their graph calculations are accurate until I get to measurements.
I think this is about as far as I can go until this weekend (which is when I start work on the cabinets/baffles), right? I hope this looks OK for this stage of the process. As always, help and advice is most appreciated.
Thanks,
Saurav
Schematic, unsmoothed response, 1/3 octave smoothed response with individual driver + network responses added in:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Not as smooth as before, and I also lost about a dB of overall efficiency, but the tweeter is much more attenuated at the lower frequencies. Also, if those breakup mode peaks aren't quite as bad in real life as they are in my SPL trace, then that will help this.
For some reason, calculating the zobel using PCDC.xls (uses Re and Le) and inserting that manually gave me better results than using Speaker Workshop's internal calculators (which always put the C at 0F). Maybe I should try PCDC for the component values too, instead of using Speaker Workshop to do the math. For now, I'll assume their graph calculations are accurate until I get to measurements.
I think this is about as far as I can go until this weekend (which is when I start work on the cabinets/baffles), right? I hope this looks OK for this stage of the process. As always, help and advice is most appreciated.
Thanks,
Saurav
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.