I'm sure I'm stirring up a hornet's nest, but I can't resist....
Several months ago I found the schematics for the KSA100 and promptly decided to build an HSPICE model using Level3 device models. Interestingly, several sub-optimum characteristics immediately became obvious, at least if distortion, pulse response, and TIM were used as figures of merit.
The first is the zener diode-only voltage regulator for the differential front end. Lacking true current sources and using only a resistor, the front end is subject to noise injection from the +/-39V supplies. When I simulated with a realistic power supply ripple, I saw 120 Hz harmonics in the output. A good first order fix is to add a bypass capacitor. A better solution would be to use voltage regulators (LM317/LM337 or the like). Another improvement is to replace the 12.1K resistors with constant current sources. (Consider using the the Vishay J500 series).
The second area of potential improvement is the psuedo-cascode VAS stage. I refer to it a pseudo-cascode because the combination of 1.00 and 1.50K resistors is only an approximation to an ideal cascode stage, where the bias voltage is independent of drive or loading. If one replaces the resistors with the combination of a constant current source and a 2.50V bandgap reference, then it is possible to achieve a near-ideal cascode configuration. Another improvement is to increase the idle current through the cascode stages by approx 2x. This can be done by adjusting the resistor values in the VN0210/VP0210 MOS stage and the 47.5 ohm cascode stage emitter resistors downward. More current yields better drive characteristics, especially when driving a capacitive load.
One final improvement is to recognize that, while very good at driving a large voltage swing, a cascode state is quite intolerant of output loading. This issue can be improved by replacing the BJT output stages with a lateral MOSFET devices. In this case, I am using four each of the BUZ901D/906D devices as outputs, driven by a pre-output state of BUZ901/906. The pre-output stage serves to reduce the Cgs load by approx 8x, and improves transient response as well as high frequency distortion -- again, because it places less of a capacitive load on the cascode stage's output.
I evaluated both BJT and MOS devices and selected the latter based on distortion vs. idle current. In a common emitter output stage, BJTs require substantial idle current to remain in their linear region. For example, an idle current of ~3A is required for minimum distortion and clean transient response. By contrast, lateral MOSFETS achieved the same characteristics with <150 ma idle current. I am not sure why this is, but if one examines the I/V transfer characteristics for the two, the BJTs have a sharper I/V knee at approx 0.6V. In other words, lateral MOSFETS have a more linear response near the V=0 region.
Right now, I am laying out the PCBs and building the chassis. I have also put in an order to Plitron for custom built toroidal transformers, where they are potted in metal enclosures with a rubberized elastomer compound to reduce acoustic vibration.
I'll post additional information as it becomes available.
Several months ago I found the schematics for the KSA100 and promptly decided to build an HSPICE model using Level3 device models. Interestingly, several sub-optimum characteristics immediately became obvious, at least if distortion, pulse response, and TIM were used as figures of merit.
The first is the zener diode-only voltage regulator for the differential front end. Lacking true current sources and using only a resistor, the front end is subject to noise injection from the +/-39V supplies. When I simulated with a realistic power supply ripple, I saw 120 Hz harmonics in the output. A good first order fix is to add a bypass capacitor. A better solution would be to use voltage regulators (LM317/LM337 or the like). Another improvement is to replace the 12.1K resistors with constant current sources. (Consider using the the Vishay J500 series).
The second area of potential improvement is the psuedo-cascode VAS stage. I refer to it a pseudo-cascode because the combination of 1.00 and 1.50K resistors is only an approximation to an ideal cascode stage, where the bias voltage is independent of drive or loading. If one replaces the resistors with the combination of a constant current source and a 2.50V bandgap reference, then it is possible to achieve a near-ideal cascode configuration. Another improvement is to increase the idle current through the cascode stages by approx 2x. This can be done by adjusting the resistor values in the VN0210/VP0210 MOS stage and the 47.5 ohm cascode stage emitter resistors downward. More current yields better drive characteristics, especially when driving a capacitive load.
One final improvement is to recognize that, while very good at driving a large voltage swing, a cascode state is quite intolerant of output loading. This issue can be improved by replacing the BJT output stages with a lateral MOSFET devices. In this case, I am using four each of the BUZ901D/906D devices as outputs, driven by a pre-output state of BUZ901/906. The pre-output stage serves to reduce the Cgs load by approx 8x, and improves transient response as well as high frequency distortion -- again, because it places less of a capacitive load on the cascode stage's output.
I evaluated both BJT and MOS devices and selected the latter based on distortion vs. idle current. In a common emitter output stage, BJTs require substantial idle current to remain in their linear region. For example, an idle current of ~3A is required for minimum distortion and clean transient response. By contrast, lateral MOSFETS achieved the same characteristics with <150 ma idle current. I am not sure why this is, but if one examines the I/V transfer characteristics for the two, the BJTs have a sharper I/V knee at approx 0.6V. In other words, lateral MOSFETS have a more linear response near the V=0 region.
Right now, I am laying out the PCBs and building the chassis. I have also put in an order to Plitron for custom built toroidal transformers, where they are potted in metal enclosures with a rubberized elastomer compound to reduce acoustic vibration.
I'll post additional information as it becomes available.
Hi JCM,
BY DEFINITION, any KRELL or KRELL KLONE must use bipolar outputs! The "K" in the KRELL logo is a graphic NPN bipolar transistor.
There are schematics on this site for the Krell KSA160 which have improved circuits and may be a good starting point for a KSA100 follow-up. The new multi-emiter power bipolars from Sanken and On-Semi have solved many design issues and should be used for any comparison to MOSFETs.
Bensen also had a MOSFET output thread that could be extended:
Complementary diff input with JFET and BJT cascode
Fully complementary FET-in FET-out design
BY DEFINITION, any KRELL or KRELL KLONE must use bipolar outputs! The "K" in the KRELL logo is a graphic NPN bipolar transistor.
There are schematics on this site for the Krell KSA160 which have improved circuits and may be a good starting point for a KSA100 follow-up. The new multi-emiter power bipolars from Sanken and On-Semi have solved many design issues and should be used for any comparison to MOSFETs.
Bensen also had a MOSFET output thread that could be extended:
Complementary diff input with JFET and BJT cascode
Fully complementary FET-in FET-out design
Thank you for clearing up the Krell nomenclature distinction.
I'll take a look at the Bensen thread.
Also, I simulated with both the original Krell output BJTs as well as the new On-Semi BJTs. The simulation results, as far as distortion, transient response, etc, were comparable.
From my understanding, the perforated emitter technology was developed to minimize secondary breakdown problems, and I believe that On Semi's perforated emitter technology is functionally similar to the multi-emitter technology you refer to in the Sanken devices.
I'll take another look at the On Semi Vbe/Ic transfer characteristics, but will be surprised if they don't exhibit a 0-600 mV range where Ic approx =0, then rises exponentially. By contrast, the BUZ901/906 start to conduct at aprox 300 mv, and their Vgs/Is slope exhibits less curvature (i.e., is more linear).
I'll take a look at the Bensen thread.
Also, I simulated with both the original Krell output BJTs as well as the new On-Semi BJTs. The simulation results, as far as distortion, transient response, etc, were comparable.
From my understanding, the perforated emitter technology was developed to minimize secondary breakdown problems, and I believe that On Semi's perforated emitter technology is functionally similar to the multi-emitter technology you refer to in the Sanken devices.
I'll take another look at the On Semi Vbe/Ic transfer characteristics, but will be surprised if they don't exhibit a 0-600 mV range where Ic approx =0, then rises exponentially. By contrast, the BUZ901/906 start to conduct at aprox 300 mv, and their Vgs/Is slope exhibits less curvature (i.e., is more linear).
courage said:
search for "Reverse engineering Krell KMA 160" There are also websites mentioned in some of the KRELL threads on this site with KRELL manuals that have schematics.
The On-Semi SPICE BJT models are not very accurate. If you search this site you will find much improved models for
MJL3281
MJL1302
Well worth the effort to adapt to your simulator.
I use Sanken 2SA1215Y 2SC2921Y to drive 20 Amps Class A into 0.15 ohm ribbons.
There is a long sticky thread on bipolar vs. mosfet outputs that has some good data.
Hi lineSource,
Can you put the schematics of the class A amp that you use with ribbons at 0.15 ohm ?
Or contact me at serraniruben@hotmail.com
Thanks!
Rubén
Can you put the schematics of the class A amp that you use with ribbons at 0.15 ohm ?
Or contact me at serraniruben@hotmail.com
Thanks!
Rubén
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.