needs a fair bit more spelled out about assumptions, models - because without something else going on several statements/conclusions are in general false
the Sallen-Key bash starting off is naive - they can be designed to work quite well for audio, Q that are encountered - few 10s of ppm smooth/low order distortion from op amp input common mode V Z nonlinearity is a non-issue audibly
the Sallen-Key bash starting off is naive - they can be designed to work quite well for audio, Q that are encountered - few 10s of ppm smooth/low order distortion from op amp input common mode V Z nonlinearity is a non-issue audibly
The original post was not intended to open a discussion, only to demonstrate the characteristics of various xover implementations as a starting point and expose graphically many common misconceptions. There are no assumptions made in the summed response analysis of the various order xovers - it is what it is.
It was anticipated that novice DIY audiophiles might appreciate an accurate and concise analysis.
Last edited:
I would think the first thing novice DIY audiophiles should be aware of is that textbook filters/XOs are generally useless, and none of the professionals I know (I do active filters for speaker in my day job) ever uses them.
You work your way backwards from the drivers instead, measuring their in-situ behaviour incl. off-axis, and derive/pick the response (mag and phase of course) you want to work with from that data set. Then you divide that response, numerically, by your chosen acoustical target (which may be textbook but doesn't need to be necessarily) and yield the required transfer function for the filter which can be fed into an convolver-based emulator.
This result is then split in minimum phase and excess phase portions and from the latter you'll know wether you need additional time-of-flight corrections (driver offsets) and/or allpass/or lowpass sections to get that delays and phase shifts (only feasible to "delay" lower freqs that way, that is).
The whole process is iterated until results in listening tests are satisfying, after that the very last step is synthesizing the actual electronic implementation with the goal of low parts count and good parameter variance immunity, ususally with a good part of intuition and experience but also with the help of an optimizer.
I usually end up with very non-textbook filters which typically incorporate shelves/dips/peaks with a filter block. Sallen-Key is excellent for this, btw. and with good opamps (and tracking supplies if you need best measured perfomance) their distortion issues are effectively non-existent.
You work your way backwards from the drivers instead, measuring their in-situ behaviour incl. off-axis, and derive/pick the response (mag and phase of course) you want to work with from that data set. Then you divide that response, numerically, by your chosen acoustical target (which may be textbook but doesn't need to be necessarily) and yield the required transfer function for the filter which can be fed into an convolver-based emulator.
This result is then split in minimum phase and excess phase portions and from the latter you'll know wether you need additional time-of-flight corrections (driver offsets) and/or allpass/or lowpass sections to get that delays and phase shifts (only feasible to "delay" lower freqs that way, that is).
The whole process is iterated until results in listening tests are satisfying, after that the very last step is synthesizing the actual electronic implementation with the goal of low parts count and good parameter variance immunity, ususally with a good part of intuition and experience but also with the help of an optimizer.
I usually end up with very non-textbook filters which typically incorporate shelves/dips/peaks with a filter block. Sallen-Key is excellent for this, btw. and with good opamps (and tracking supplies if you need best measured perfomance) their distortion issues are effectively non-existent.
This issue was addressed in Post # 37.
It would seem that you have a vested interest in perpetuating the myth that exotic drivers with bizarre xovers are needed to provide exceptional fidelity.
Or maybe you just enjoy the chase.
Last edited:
The behavior of the electrical summed response of multi-way xovers has nothing to do with driver behavior. Before one can attempt to choose an xover solution, this exact electrical nature should be understood. Browsing through vast posts on this site reveals a general lack of this knowledge.
I'm not really sure what the point of this thread is. I guess it's just to educate the ignorant masses here on DIYaudio.com?
JrmEng has made a grand total of 29 posts on DIYaudio.com....every one of them in this thread. He has no interest in discussing his conclusions and (apparently) no interest in the rest of DIYaudio.com.
Clearly there's some sort of underlying agenda here, but whatever it is, it's a surreal way to promote it.
I think just about everyone who had been following this thread is no longer.
Dave.
JrmEng has made a grand total of 29 posts on DIYaudio.com....every one of them in this thread. He has no interest in discussing his conclusions and (apparently) no interest in the rest of DIYaudio.com.
Clearly there's some sort of underlying agenda here, but whatever it is, it's a surreal way to promote it.
I think just about everyone who had been following this thread is no longer.
Dave.
I am using a 3-way 3rd order crossover system and completely agree with most of his observations, having also experimented with 4th, 5th, and 8th orders. Also, there is a distinct change in the sound signature when phase across the audio band becomes too great, as in going to higher than 3rd order in a 3 or 4-way. This has nothing to do with the group delay effect, which is a gross manifestation of very-very large phase shift.
As to SPICE models, he is also right that it makes no difference in frequency response analysis which opamp model is used (I routinely use SPICE in my profession). I have also discovered that the new transistor models that are generated from the data sheet are not accurate and make them useless for distortion analysis. And opamp models only implement minimum internal transistor parameters making them also useless for distortion analysis – they always yield ridiculously low values.
Transient-perfect crossovers
Clarified transient-perfect nature of 1st order 4-way and 2nd and 3rd order asymmetrical 2-way.
http://auratron.co.nf.Cascade.htm
Clarified transient-perfect nature of 1st order 4-way and 2nd and 3rd order asymmetrical 2-way.
http://auratron.co.nf.Cascade.htm
Mechanical Time Alignment
I have read this section through twice.
I don't understand what mechanical time alignment is?
Can you explain in a different way so that I may understand?
I have read this section through twice.
I don't understand what mechanical time alignment is?
Can you explain in a different way so that I may understand?
http://auratron.co.nf/Cascade.htm#ta
Mechanical Time Alignment
Consider a 2-way system with 2 drivers. Accurate time alignment is achieved when the arrival time of the energy from each driver reaches the listener at the same time. For typical woofer / tweeter mounted flush on a baffle, the energy will not arrive at the same time because the actual ACOUSTIC position of the drivers is not the same and is not at the front of the baffle. Mis-alignment will result in a peak or dip in frequency response around xover, depending on xover topology.
Accurate mechanical time-alignment is achieved if a method can be devised to detect the arrival time of the energy from the 2 drivers and adjust one forward or back until the arrival times are equal. Of course, this means the drivers will have to be mounted on separate baffles such that the position can be adjusted independently. The energy-time or impulse response is typically used for this determination.
Another method is to adjust for flat frequency response in the octave at, above, and below xover, but this method will only work if the drivers or xover have no anomalies.
It is possible, and sometimes very likely, that adjusting for flat response will not result in correct time-alignment as described above. If the listener is not sensitive to time-alignment error, then this method will provide satisfactory results, and would be the optimum solution for those individuals.
Time-alignment error also results in an effective relative phase shift in the xover region and many methods have been proposed to introduce an electrical network, mis-align the xover, play polarity games in an attempt to correct for this phase error. While it is possible to achieve almost-flat response at xover using these methods, the results will not be sonically equal to that achieved by true mechanical alignment, especially in those sensitive to time-alignment error in spite of frequency response.
Our preferred method is to measure the unwrapped phase of the system and adjust the driver positions until it's phase matches that of the unwrapped phase of the xover summed response. This method removes frequency response errors from the measurement.
The SPICE simulations show unwrapped phase for the various xover topologies.
When making the measurement, first remove the initial time delay of drivers to microphone, the starting phase should be nearly zero; if not reverse polarity of the microphone (or somewhere in the measurement chain). Then the unwrapped phase of the drivers + xover can be matched to that of the xover unwrapped phase by adjusting driver positions. Actually, without looking at the whole spectrum at once, a few points in the phase of the xover summed response can be determined (measured (preferred) or calculated) and used to set phase of measured system.
It is most interesting to compare the results of the impulse method, flat response method, and phase method. If the phase method results in response peaks / dips, then simple EQ topology corrections can be added in the xover.
After 35+ years of experimenting with these techniques, we have concluded that the "best" sound results from the phase method, "best" meaning most sonically neutral with greatest sense of sound stage dimension.
Consider a 2-way system with 2 drivers. Accurate time alignment is achieved when the arrival time of the energy from each driver reaches the listener at the same time. For typical woofer / tweeter mounted flush on a baffle, the energy will not arrive at the same time because the actual ACOUSTIC position of the drivers is not the same and is not at the front of the baffle. Mis-alignment will result in a peak or dip in frequency response around xover, depending on xover topology.
Accurate mechanical time-alignment is achieved if a method can be devised to detect the arrival time of the energy from the 2 drivers and adjust one forward or back until the arrival times are equal. Of course, this means the drivers will have to be mounted on separate baffles such that the position can be adjusted independently. The energy-time or impulse response is typically used for this determination.
Another method is to adjust for flat frequency response in the octave at, above, and below xover, but this method will only work if the drivers or xover have no anomalies.
It is possible, and sometimes very likely, that adjusting for flat response will not result in correct time-alignment as described above. If the listener is not sensitive to time-alignment error, then this method will provide satisfactory results, and would be the optimum solution for those individuals.
Time-alignment error also results in an effective relative phase shift in the xover region and many methods have been proposed to introduce an electrical network, mis-align the xover, play polarity games in an attempt to correct for this phase error. While it is possible to achieve almost-flat response at xover using these methods, the results will not be sonically equal to that achieved by true mechanical alignment, especially in those sensitive to time-alignment error in spite of frequency response.
Our preferred method is to measure the unwrapped phase of the system and adjust the driver positions until it's phase matches that of the unwrapped phase of the xover summed response. This method removes frequency response errors from the measurement.
The SPICE simulations show unwrapped phase for the various xover topologies.
When making the measurement, first remove the initial time delay of drivers to microphone, the starting phase should be nearly zero; if not reverse polarity of the microphone (or somewhere in the measurement chain). Then the unwrapped phase of the drivers + xover can be matched to that of the xover unwrapped phase by adjusting driver positions. Actually, without looking at the whole spectrum at once, a few points in the phase of the xover summed response can be determined (measured (preferred) or calculated) and used to set phase of measured system.
It is most interesting to compare the results of the impulse method, flat response method, and phase method. If the phase method results in response peaks / dips, then simple EQ topology corrections can be added in the xover.
After 35+ years of experimenting with these techniques, we have concluded that the "best" sound results from the phase method, "best" meaning most sonically neutral with greatest sense of sound stage dimension.
"Phase Method" Correction
The paragraph should read:
When making the measurement, first remove the initial time delay of drivers to microphone; the starting phase should then be equal to the xover sum at the starting f (Δf) of the measurement system; if not reverse polarity of the microphone (or somewhere in the measurement chain). Then the unwrapped phase of the drivers + xover can be matched to that of the xover unwrapped phase alone by adjusting driver positions. Of most importance is that the slope of the phase be as close as possible to that of the xover sum. Actually, without looking at the whole spectrum at once, a few points in the phase of the xover summed response can be determined (measured (preferred) or calculated) and used to set phase of measured system.
The paragraph should read:
When making the measurement, first remove the initial time delay of drivers to microphone; the starting phase should then be equal to the xover sum at the starting f (Δf) of the measurement system; if not reverse polarity of the microphone (or somewhere in the measurement chain). Then the unwrapped phase of the drivers + xover can be matched to that of the xover unwrapped phase alone by adjusting driver positions. Of most importance is that the slope of the phase be as close as possible to that of the xover sum. Actually, without looking at the whole spectrum at once, a few points in the phase of the xover summed response can be determined (measured (preferred) or calculated) and used to set phase of measured system.
Inverting vs. Non-inverting Feedback Amplifiers
In order for the full advantage of feedback to be realized, the feedback signal and the input signal must follow the same path through the circuit, otherwise errors will be introduced.
If non-differential input is used, feedback is applied to emitter of input transistor, introducing a B-E discrepancy. When using differential in (circuit) non-inverting mode, there is a 2 x B-E discrepancy.
With the exception of DiFETs (OPA627/637), all opamps will have an additional source of error due to reverse bias diode-substrate leakage that exists for all components on the die (Substrate is connected to V-), and input capacitance-to-substrate modulation.
Substrate leakage artifacts are the root cause of the "Opamp Sound".
The solution is differential input where input node and feedback node are identical. This will also ensure the application of feedback does not introduce any new distortions.
Makes no difference if op amp, discrete transistor, or tube implementation.
In order for the full advantage of feedback to be realized, the feedback signal and the input signal must follow the same path through the circuit, otherwise errors will be introduced.
If non-differential input is used, feedback is applied to emitter of input transistor, introducing a B-E discrepancy. When using differential in (circuit) non-inverting mode, there is a 2 x B-E discrepancy.
With the exception of DiFETs (OPA627/637), all opamps will have an additional source of error due to reverse bias diode-substrate leakage that exists for all components on the die (Substrate is connected to V-), and input capacitance-to-substrate modulation.
Substrate leakage artifacts are the root cause of the "Opamp Sound".
The solution is differential input where input node and feedback node are identical. This will also ensure the application of feedback does not introduce any new distortions.
Makes no difference if op amp, discrete transistor, or tube implementation.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.