• LC-LC 4th order low pass filter
  
• Replace the AD797 (U2) with an OP27. AD797's are bad - do not use them anywhere for any reason. The OP27 is slower and has a 3x worse input noise but doesn't compromise the bandwidth or noise performance of the PMC by any significant amount. The rule is: use OP27 everywhere unless you have a very good reason why not.
  
• There is no 'H1' jumper on board. R9 is 90.9 Ohms and R2=900 Ohms so that the U2 stage has a gain of 10.
  
• Cut a trace and inserted a 500 Ohm resistor between U2-pin6 and U5A-pin2 (the AD602). The AD602 has a 100 Ohm input impedance which cannot be driven without limiting by the AD797 or the OP27. The 500 Ohm resistor makes it a driveable load for low level signals which is all that should be there since its the error point of the servo. it also becomes a 6:1 voltage divider. Since the AD602 has a fixed output voltage noise of 100 nV/rHz, this will limit the noise performance if the VGA gain is less than 20 dB, but whatever.
  
• R11 7.87k -> 1.74k, R12 = 78.7k -> 700k. This increases the high frequency gain of that stage by 7.87/1.74 = 4.5 and lowers the low frequency pole from 2 to 0.2 Hz to give the PMC some more staying power at DC. The loop shape is now 1/f^2 in the 9-480 Hz band and so the phase dips enough to make it almost conditionally stable, but not quite.
  
• C26 changed from ??? + a 30 pF trim cap into a fixed NNN pF cap to set the notch frequency for the 14.5 kHz body mode that we measured. Once our brick configuration is more settled we can increase the Q of this notch from small to big.
  
• Grounded pin 5 of U14 & U15 (AD620). These have sometimes been used as "differential" drivers in LIGO by connecting this reference voltage pin to the remote ground of the next board. This has always lead to insidious oscillation and noise. This beauty also has an output noise of 100 nV/rHz. Just never use this chip if you can help it; we can make true differential drivers - we have the technology.
  

[C1:PEM-ACC_MC1_X]
chnnum=15014
gain = 10

etc.

Summary:
Since we reduced the integration time of the particle counter by a factor of 10, we had to add a gain of 10
to the EPICS channels C1:PEM-count_full and C1:PEM-count_half.
I asked Alex to change it and he did it. I forgot to ask him to change the gain of C1:PEM-count_half. So now only
C1:PEM-count_full has x10 gain.

Detail:
C1:PEM-count_full and C1:PEM-count_half are 'Soft Channel' records in the database (Pcount.db). The values are actually
written into the VAL fields directly by an SNL code Particle.o.
Particle.o reads data from the RS-232C port, to which the particle counter is connected. Then it parses the data and put values
into relevant EPICS channels using channel access. This means we cannot change the gain of the channels by modifying the
database file. For example, ASLO field does not have any effect when the value is directly written into the VAL field.
We had to modify the SNL code. Alex modified Particle.st and the new SNL object file is Particle_x10.o sitting in 
/cvs/cds/caltech/target/c1psl/. I modified seq.load so that c1psl loads Particle_x10.o when rebooted.
The source code for the old Particle.st can be found on lesath.ligo.caltech.edu in
/export/CDS/d/epics/apple/Caltech/40mVac/40mVacScipe/dev/src
I asked Alex to disclose the location of the source of the new code.
In order to compile the SNL code into an object file for Motorola CPU by ourselves, we have to call Dave Barker at LHO.

At the request of Steve, I modified the HIGH value of C1:PSL-FSS_RMTEMP from 21.27 to 23.0.
The HIHI is set to 23.50.

SR620                Marconi
                     199178070
165981524            165981725
                     132785380
                      33196345

sudo yum install grace