iESR=in circuit ESR
ESR=out of circuit ESR
WV=working voltage of a functional unit
D-Link DWL-P200 POE Adapter
A/N: PCPE1004-000-00300 P/N:PE1004 Rev.00 START:2004/05/06
D1, D2: 16 SMD Diode 2.85x1.70
D6: WF SMD Diode 1.72x1.21
R9, R13: 000 SMD Resistor
D7: SK56 SMD Diode 6.74x5.84 Freewheel
Q3: OnSemi NT2955G NTP2955G TO-252 P-Channel MOSFET
Q1, Q5: SAN SOT-23 SMD Transistor 2.87x1.20
Q2: 1AM SOT-23 SMD Transistor 2.90x1.30
Q4: 2L SOT-23 SMD Transistor 2.93x1.27
U3: TI TL5001C SOIC-8 PWM Control C25:48v -> C19:12v/5v
C25: Stone 100uf 63v 8x14 WV=47.6v iESR=0.38 ESR=0.33
C19: Stone 100uf 25v 8x10 WV=5v/12v iESR=0.27 ESR=0.34
D-Link DWL-700AP Access Point HW Ver: A1 FCCID:KA2DWL700AP1-A1 331DC 2A 94V-0 WA254D-02-B
UX1: AIC1084CM TO-263 LDO Adjustable Linear Regulator biased to put out 3.3v
UX2: Promax-Johnton PJ1117CW-2.5 SOT-223 1A LDO 2.5 Linear Regulator
U2: Sipex SP3243ECA Intelligent +3.0V to +5.5V RS-232 Transceivers
U6: ADMtek ADM8628 ARM7TDMI Microrprocessor Ethernet to Wifi bridge with AES WEP TKIP HTTP SNMP
U10: Hynix HY57V64322OCT-7 4 Banks x 512K x 32Bit Synchronous DRAM
TE2: GTS FC-618SM 10/100 BASE-T SMT TRANSFORMER
D2: Jumper
U1: AFAS SMD Transistor 2.93x1.24
CX2, CX1, WV=5v iESR=0.26
C7, WV=3.3v iESR=0.16
C9: Fuhjyyu TNR 220uf 16v 6x11 WV=2.5v iESR=0.42
C87, C109, WV=2.5v iESR=0.42
C94, C110, C136, C138: Fuhjyyu TNR 10uf 50v 5x11 WV=3.3v iESR=0.16
Empty spots:
C13, C5, C58: WV=3.3v
C8: WV=2.5v
Two of these combinations run for about 2 weeks then disappear. Other brand access points with integrated POE on the same midspan are working well.
The only switching regulator in the combo is 5v/12v in the POE power supply. The access point only has linear regulators 3.3v and 2.5v.
Scope Pictures:
DSCF3107 48v at POE-C25 with the original Stone capacitor
DSCF3108 48v at POE-C25 with the original Stone capacitor and a Nichicon PW 220uf 63v ESR=0.09 long leads in parallel. Major improvement.
DSCF3113 5v at POE-C19 with the original Stone capacitor. Adding the Nichicon PW into 48v circuit does not change this waveform.
DSCF3114 5v at POE-C19 with the original Stone capacitor and a Rubycon MCZ 1800uf 6.3v ESR=0.01 long leads in parallel.
The fuzziness of these pictures may be due to being so close to the noise floor of the scope or because of the tenancy of SMPS circuits to overlay repeating sets of similar waveforms that trigger at different points along the wave because of the difference in voltage. This is easy to see with SMPS circuits that only have a few similar waveforms that paint in a clear pattern instead of fuzz.
Scope Pictures:
DSCF3115 5v at AP-UX1 to each linear regulator. Adding the Rubycon 1800uf into this 5v supply cuts the 5v ripple in half.
DSCF3116 Output of each AP linear regulator. Both regulator outputs look the same. Adding the Rubycon 1800uf into the 5v circuit reduces the white noise on the 3.3v and 2.5v rails but does not reduce the ripple.
I see only two failure points that could be improved. Adding the Nichicon PW capacitor into the 48v circuit balances out the waveform. This doesn't much matter over 4 feet of POE supply cable in the test setup but could be quite substantial over 100' of cable in the working environment. A better capacitor for POE supply C28 could be helpful in stabilizing the 48v supply over long cable run. A better capacitor in the access point 5v supply does not substantially improve the outputs of the linear regulators but could help whatever parts that run at 5v.
New caps:
POE-C25: Nichicon PW 220uf 63v ESR=0.09 (long leads)
AP-CX1: Rubycon MCZ 820uf 6.3v ESR=0.01
Scope Pictures:
DSCF3117 48v at POE-C25 with Nichicon soldered in with 1/4 inch extra leads. DISASTER!
DSCF3119 48v at POE-C25 with Nichicon soldered in with 1/4 inch extra leads and original Stone 100uf 63v placed in parallel. Fixed!
DSCF3120 48v at POE-C25 with Nichicon soldered in with short leads. Excellent but won't fit.
The Nichicon is too big for the case so I left on an extra 1/4 inch of leads and bent it over. This has been done in other products so it seemed perfectly reasonable. Unfortunately it caused a disaster on the scope. I melted the solder and pushed the Nichicon almost flush to the board which made the waveform better than with the original Stone capacitor. Surprised that 1/2 inch extra wire caused total disaster, I wanted to demonstrate that lead length was the problem and not the soldering or scope work. I desoldered both leads so I could slide the capacitor in and out with the scope attached. It was a challenge to keep the capacitor connected but extending the lead length really did grow the ring that much. As nice as it would be to use the the big Nichicon cap it just won't work because the leads can't be short enough and still fit. When I was first testing the Nichicon in parallel with the Stone in the 48v circuit I tried hooking the Nichicon negative lead to a convenient ground by a 16 inch alligator clip and touching the positive lead to the solder joint. The waveform improvement was almost imperceptible. Placing the Nichicon that worked so well flush to the board behind 18 inches of wire made it almost nonexistent to the circuit.
I paw through the cap box to find a cap that will fit flush in the case with good ESR.
POE-C25: Teapo SEK 100uf 50v ESR=0.13
Scope Pictures:
DSCF3121 48v at POE-C25 with Teapo. Better than the larger Nichicon 220uf cap, possibly because it fits flush making the leads even shorter. The Nichicon leads were too far apart and couldn't be pushed all the way flush to the board. The stone cap in parallel helps but not a lot. This suggests that more uF would be better.
DSCF3123 5v at POE-C19 with Teapo in POE-C25. Same as before which indicates that the output of the SMPS isn't affected much by the input until it gets really bad.
Alas! Too much testing has burned the access point out. Too much accidental shorting and unreliable power from sliding the Nichicon capacitor in and out. The Ethernet port works but the wifi bridge program does not start. LED3 no longer lights. 5v, 3.3v, and 2.5v in the access point are still good and the ripple is below the scope's noise floor. But that's not enough damage for one night. In goes another cap.
POE-C19: Panasonic FJ 470uf 16v ESR=0.02
Scope Pictures:
DSCF3124 5v at POE-C19 with Panasonic. A bit of ripple but very close to the noise floor of the scope.
Replacing CX1 in the access point for 5v to the linear regulators was probably unnecessary. Not much noise comes back out of linear regulators so there's nothing to filter here. The ripple started at the SMPS in the POE adapter and that's the best place to stop it. Since the 5v output was not affected by waveform improvements on the 48v input replacing the cap on 48v may not have been necessary. Like so much other SMPS powered networking equipment, this duo may have only needed the SMPS output cap improved. A better cap in the 48v circuit just makes the POE power supply more reliable on long cable runs. Note that both original POE power supply Stone caps have better than average ESR. They just weren't good enough to make the connected equipment reliable.
This access point is dead so can't be deployed. The POE power supply is now in great shape and I can go on site and swap it onto the unmodified access point with no down time.
Update: The improved POE adapter and the unmodified access point has been running for a few months with no down time.
ESR=out of circuit ESR
WV=working voltage of a functional unit
D-Link DWL-P200 POE Adapter
A/N: PCPE1004-000-00300 P/N:PE1004 Rev.00 START:2004/05/06
D1, D2: 16 SMD Diode 2.85x1.70
D6: WF SMD Diode 1.72x1.21
R9, R13: 000 SMD Resistor
D7: SK56 SMD Diode 6.74x5.84 Freewheel
Q3: OnSemi NT2955G NTP2955G TO-252 P-Channel MOSFET
Q1, Q5: SAN SOT-23 SMD Transistor 2.87x1.20
Q2: 1AM SOT-23 SMD Transistor 2.90x1.30
Q4: 2L SOT-23 SMD Transistor 2.93x1.27
U3: TI TL5001C SOIC-8 PWM Control C25:48v -> C19:12v/5v
C25: Stone 100uf 63v 8x14 WV=47.6v iESR=0.38 ESR=0.33
C19: Stone 100uf 25v 8x10 WV=5v/12v iESR=0.27 ESR=0.34
D-Link DWL-700AP Access Point HW Ver: A1 FCCID:KA2DWL700AP1-A1 331DC 2A 94V-0 WA254D-02-B
UX1: AIC1084CM TO-263 LDO Adjustable Linear Regulator biased to put out 3.3v
UX2: Promax-Johnton PJ1117CW-2.5 SOT-223 1A LDO 2.5 Linear Regulator
U2: Sipex SP3243ECA Intelligent +3.0V to +5.5V RS-232 Transceivers
U6: ADMtek ADM8628 ARM7TDMI Microrprocessor Ethernet to Wifi bridge with AES WEP TKIP HTTP SNMP
U10: Hynix HY57V64322OCT-7 4 Banks x 512K x 32Bit Synchronous DRAM
TE2: GTS FC-618SM 10/100 BASE-T SMT TRANSFORMER
D2: Jumper
U1: AFAS SMD Transistor 2.93x1.24
CX2, CX1, WV=5v iESR=0.26
C7, WV=3.3v iESR=0.16
C9: Fuhjyyu TNR 220uf 16v 6x11 WV=2.5v iESR=0.42
C87, C109, WV=2.5v iESR=0.42
C94, C110, C136, C138: Fuhjyyu TNR 10uf 50v 5x11 WV=3.3v iESR=0.16
Empty spots:
C13, C5, C58: WV=3.3v
C8: WV=2.5v
Two of these combinations run for about 2 weeks then disappear. Other brand access points with integrated POE on the same midspan are working well.
The only switching regulator in the combo is 5v/12v in the POE power supply. The access point only has linear regulators 3.3v and 2.5v.
Scope Pictures:
DSCF3107 48v at POE-C25 with the original Stone capacitor
DSCF3108 48v at POE-C25 with the original Stone capacitor and a Nichicon PW 220uf 63v ESR=0.09 long leads in parallel. Major improvement.
DSCF3113 5v at POE-C19 with the original Stone capacitor. Adding the Nichicon PW into 48v circuit does not change this waveform.
DSCF3114 5v at POE-C19 with the original Stone capacitor and a Rubycon MCZ 1800uf 6.3v ESR=0.01 long leads in parallel.
The fuzziness of these pictures may be due to being so close to the noise floor of the scope or because of the tenancy of SMPS circuits to overlay repeating sets of similar waveforms that trigger at different points along the wave because of the difference in voltage. This is easy to see with SMPS circuits that only have a few similar waveforms that paint in a clear pattern instead of fuzz.
Scope Pictures:
DSCF3115 5v at AP-UX1 to each linear regulator. Adding the Rubycon 1800uf into this 5v supply cuts the 5v ripple in half.
DSCF3116 Output of each AP linear regulator. Both regulator outputs look the same. Adding the Rubycon 1800uf into the 5v circuit reduces the white noise on the 3.3v and 2.5v rails but does not reduce the ripple.
I see only two failure points that could be improved. Adding the Nichicon PW capacitor into the 48v circuit balances out the waveform. This doesn't much matter over 4 feet of POE supply cable in the test setup but could be quite substantial over 100' of cable in the working environment. A better capacitor for POE supply C28 could be helpful in stabilizing the 48v supply over long cable run. A better capacitor in the access point 5v supply does not substantially improve the outputs of the linear regulators but could help whatever parts that run at 5v.
New caps:
POE-C25: Nichicon PW 220uf 63v ESR=0.09 (long leads)
AP-CX1: Rubycon MCZ 820uf 6.3v ESR=0.01
Scope Pictures:
DSCF3117 48v at POE-C25 with Nichicon soldered in with 1/4 inch extra leads. DISASTER!
DSCF3119 48v at POE-C25 with Nichicon soldered in with 1/4 inch extra leads and original Stone 100uf 63v placed in parallel. Fixed!
DSCF3120 48v at POE-C25 with Nichicon soldered in with short leads. Excellent but won't fit.
The Nichicon is too big for the case so I left on an extra 1/4 inch of leads and bent it over. This has been done in other products so it seemed perfectly reasonable. Unfortunately it caused a disaster on the scope. I melted the solder and pushed the Nichicon almost flush to the board which made the waveform better than with the original Stone capacitor. Surprised that 1/2 inch extra wire caused total disaster, I wanted to demonstrate that lead length was the problem and not the soldering or scope work. I desoldered both leads so I could slide the capacitor in and out with the scope attached. It was a challenge to keep the capacitor connected but extending the lead length really did grow the ring that much. As nice as it would be to use the the big Nichicon cap it just won't work because the leads can't be short enough and still fit. When I was first testing the Nichicon in parallel with the Stone in the 48v circuit I tried hooking the Nichicon negative lead to a convenient ground by a 16 inch alligator clip and touching the positive lead to the solder joint. The waveform improvement was almost imperceptible. Placing the Nichicon that worked so well flush to the board behind 18 inches of wire made it almost nonexistent to the circuit.
I paw through the cap box to find a cap that will fit flush in the case with good ESR.
POE-C25: Teapo SEK 100uf 50v ESR=0.13
Scope Pictures:
DSCF3121 48v at POE-C25 with Teapo. Better than the larger Nichicon 220uf cap, possibly because it fits flush making the leads even shorter. The Nichicon leads were too far apart and couldn't be pushed all the way flush to the board. The stone cap in parallel helps but not a lot. This suggests that more uF would be better.
DSCF3123 5v at POE-C19 with Teapo in POE-C25. Same as before which indicates that the output of the SMPS isn't affected much by the input until it gets really bad.
Alas! Too much testing has burned the access point out. Too much accidental shorting and unreliable power from sliding the Nichicon capacitor in and out. The Ethernet port works but the wifi bridge program does not start. LED3 no longer lights. 5v, 3.3v, and 2.5v in the access point are still good and the ripple is below the scope's noise floor. But that's not enough damage for one night. In goes another cap.
POE-C19: Panasonic FJ 470uf 16v ESR=0.02
Scope Pictures:
DSCF3124 5v at POE-C19 with Panasonic. A bit of ripple but very close to the noise floor of the scope.
Replacing CX1 in the access point for 5v to the linear regulators was probably unnecessary. Not much noise comes back out of linear regulators so there's nothing to filter here. The ripple started at the SMPS in the POE adapter and that's the best place to stop it. Since the 5v output was not affected by waveform improvements on the 48v input replacing the cap on 48v may not have been necessary. Like so much other SMPS powered networking equipment, this duo may have only needed the SMPS output cap improved. A better cap in the 48v circuit just makes the POE power supply more reliable on long cable runs. Note that both original POE power supply Stone caps have better than average ESR. They just weren't good enough to make the connected equipment reliable.
This access point is dead so can't be deployed. The POE power supply is now in great shape and I can go on site and swap it onto the unmodified access point with no down time.
Update: The improved POE adapter and the unmodified access point has been running for a few months with no down time.
Attached Files
