The noise levels of the output pins (pin1/pin6) are measured. Note that the measurement is done with SE. i.e. There was no common mode noise rejection.

Worked on the calibration of the RF AM Measurement Unit.

The calibration concept is as follows:

• Generate AM modulated RF output
• Measure sideband amplitude using a network anayzer (HP4395A). This gives us the SSB carrier-sideband ratio in dBc.
• Measure the output of the RF AM measurement unit with the same RF signal
• Determine the relationship between dBc(SSB) and the output Vrms.

The AM modulated signal is produced using DS345 function generator. This FG allows us to modulate
the output by giving an external modulation signal from the rear panel. In the calibration, a 1kHz signal with
the DC offset of 3V was given as the external modulation source. The output frequency and output power of
DS345 was set to be 30.2MHz (maximum of the unit) and 14.6dBm. This actually imposed the output
power of 10.346dBm. Here is the result with the modulation amplitude varied

                 RF Power measured             Monitor output
Modulation       with HP4395A (dBm)           Measure with SR785 (mVrms)
1kHz (mVpk)   Carrier    USB      LSB          MON1       MON2
   0.5        9.841    -72.621  -73.325          8.832      8.800
   1          9.99     -65.89   -65.975         17.59      17.52
   2          9.948    -60.056  -59.747         35.26      35.07
   3          9.90     -56.278  -56.33          53.04      52.9
   5          9.906    -51.798  -51.797         88.83      88.57
  10          9.892    -45.823  -45.831        177.6      177.1
  20          9.870    -39.814  -39.823        355.3      354.4
  30          9.8574   -36.294  -36.307        532.1      531.1
  50          9.8698   -31.86   -31.867        886.8      885.2
 100          9.8735   -25.843  -25.847       1772       1769
 150          9.8734   -22.316  -22.32        2656       2652
 200          9.8665   -19.819  -19.826       3542       3539
 300          9.8744   -16.295  -16.301       5313       5308

The SSB carrier sideband ratio is derived by SSB[dBc] = (USB[dBm]+LSB[dBm])/2 - Carrier[dBm]

This measurement suggests that 10^(dBc/20) and Vrms has a linear relationship. (Attachment 1)
The data points were fitted by the function y= a x.

=> 10^dBc(SSB)/20 = 108*Vrms (@10.346dBm input)

Now we want to confirm this calibration.

DS345 @30.2MHz was modulated with the DC offset + random noise. The resulting AM modulated RF was checked with the network analyzer and the RFAM detector
in order to compare the calibrated dBc/Hz curves.

A) SR785 was set to produce random noise
B) Brought 2nd DS345 just to produce the DC offset of -2.52V (Offset display -1.26V)
Those two are added (A-B) by an SR560 (DC coupling, G=+1, 50 Ohm out).
The output was fed to Ext AM in DS345(#1)

DS345(#1) was set to 30.2MHz 16dBm => The measured output power was 10.3dBm.

On the network analyzer the carrier power at 30.2MHz was 9.89dBm

Measurement 1) SR785     1.6kHz span 30mV random noise (observed flat AM noise)
Measurement 2) SR785   12.8kHz span 100mV random noise (observed flat AM noise)
Measurement 3) SR785 102.4kHz span 300mV random noise (observed cut off of the AM modulation due to the BW of DS345)

The comparison plot is attached as Attachment 2. Note that those three measurements are independent and are not supposed to match each other.
The network analyzer result and RFAM measurement unit output should agree if the calibration is correct. In fact they do agree well.

In order to check the noise level of the RFAM detector, the power and cross spectra for the same signal source
were simultaneously measured with the two RFAM detectors.

As a signal source, 37MHz OCXO using a wenzel oscillator was used. The output from the signal source
was equaly splitted by a power splitter and fed to the RFAM detector CHB(Mon1) and CHA(Mon2).

The error signal for CHB (Mon1) were monitored by an oscilloscope to find an appropriate bias value.
The bias for CHA are adjusted automatically by the slow bias servo.

The spectra were measured with two different power settings:

Low Power setting: The signal source with 6+5dB attenuation was used. This yielded 10.3dBm at the each unit input.
The calibration of the low power setting is dBc = 20*log10(Vrms/108). (See previous elog entry)

High Power setting: The signal source was used without any attenuation. This yielded 22.4dBm at the each unit input.
The calibration for the high power setting was measured upon the measurement.
SR785 was set to have 1kHz sinusoidal output with the amplitude of 10mVpk and the offset of 4.1V.
This modulation signal was fed to DS345 at 30.2MHz with 24.00dBm
The network analyzer measured the carrier and sideband power levels
30.2MHz 21.865dBm
USB    -37.047dBm
LSB    -37.080dBm  ==> -58.9285 dBc (= 0.0011313)

The RF signal was fed to the input and the signal amplitude at Mon1 and Mon2 were measured
MON1 => 505   mVrms => 446.392 Vrms/ratio
MON2 => 505.7 mVrms => 447.011 Vrms/ratio
dBc = 20*log10(Vrms/446.5).

Using the cross specrum (or coherence)of the two signals, we can infer the noise level of the detector.

Suppose there are two time-series x(t) and y(t) that contain the same signal s(t) and independent but same size of noise n(t) and m(t)

x(t) = n(t) + s(t)
y(t) = m(t) + s(t)

Since n, m, s are not correlated, PSDs of x and y are

Pxx = Pnn + Pss
Pyy = Pmm+Pss = Pnn+Pss

The coherence between x(t) and y(t) is defined by

Cxy = |Pxy|^2/Pxx/Pyy = |Pxy|^2/Pxx^2

In fact |Pxy| = Pss. Therefore

sqrt(Cxy) = Pss/Pxx

What we want to know is Pnn

Pnn = Pxx - Pss = Pxx[1 - sqrt(Cxy)]
=> Snn = sqrt(Pnn) = Sxx * sqrt[1 - sqrt(Cxy)]

This is slightly different from the case where you don't have the noise in one of the time series (e.g. feedforward cancellation or bruco)

Measurement results

Power spectra:
Mon1 and Mon2 for both input power levels exhibited the same PSD between 10Hz to 1kHz. This basically supports that the calibration for the 22dBm input (at least relative to the calibration for 10dBm input) was corrected. Abobe 1kHz and below 10Hz, some reduction of the noise by the increase of the input power was observed. From the coherence analysis, the floor level for the 10dBm input was -178, -175, -155dBc/Hz at 1kHz, 100Hz, and 10Hz, respectively. For the 22dBm input, they are improved down to -188, -182, and -167dBc/Hz at 1kHz, 100Hz, and 10Hz, respectively.

TEST Result: S1500118

Additional notes

- Checked the power supply. All voltages look quiet and stationary.

- Checked the internal RF cables too see if there is any missing shield soldering => Looked fine

- Noticed that the RFAM detector board has +/-21V for the +/-24V lines => It seems that this is nominal according to the schematic

- Noticed that the RFAM detector sensitivity were doubled fomr the other unit.
  => This is reated to the modification (E1500353) of  "Controller Board D0900761-B Change 1" (doubling the monitor output gain)

- Noticed that the transfer function of the CTRL signal on the BNC and the DAQ output.
  => This is reated to the modification (E1500353) of  "Servo Board D0900847-B Change 1"  (servo transfer function chage)
  => The measured transfer function did not agree with the prediction from the circuit constants in this document
  => From the observation of the servo board it was found that R69 was not 200Ohm but 66.5 Ohm (See attachment 1).
       This explained the measured transfer function. The actuator TF has: P 2.36, Z 120., K -1@DC (0.020@HF)

- Similarly, the TF between the CTRL port on the unit and the CTRL port on the test rig was also modified.

Noise level

Attachment 2

- The amplitude noise in dBc (SSB) was measured at the output of 27dBm. From the test sheet, the noise level with 13dBm output was also referred. From the coherence of the MON1 and MON2, the noise level was inferred. It suggests that the floor level is better than 180dBc/Hz. However, there is a 1/f like noise below 1k and is dominating the actual noise level of the RF output. Daniel suggested that we should check nonlinear downconversion from the high frequency noise due to the noise attenuator. This will be check with the coming units.

Test sheet: https://dcc.ligo.org/LIGO-E1400445

Test Result (S1500114): https://dcc.ligo.org/S1500114

Linearity of the EOM driver was tested. This test has been done on Nov 10, 2015.

- Attachment 1: Output power vs requested power. The output start to deviate from the request above 22dBm request.

- Attachment 2: Ctrl and Bias voltages vs requested power. This bias was measured with the out-of-loop channel.
The variable attenuator has the voltage range of 0~15V for 50dB~2dB attenuation.

Therefore this means that:

- The power setting gives a voltage logarithmically increased as the requested power increases. And the two power detectors are watching similar voltages.

- And the servo is properly working. The control is with in the range.

- Even when the given RF power is low, the power detectors are reporting high value. Is there any mechanism to realize such a condition???

This measurement has been done on Dec 1st, 2015.

The phase noise added by the EOM driver was tested.

The test setup is depicted in the attached PDF. The phase of the RF detector was set so that the output is close to zero crossing as much as possible with the precision of 0.5ns using a switchable delay line box. The phase to voltage conversion was checked by changing the delayline by 1ns. This gave me somewhat larger conversion factor compared to the sine wave test using an independent signal generator. This was due to the saturation of the phase detector as the LO and RF both have similar high RF level for the frequency mixer used.

The measurement has done with 1) no EOM driver involved, 2) one EOM driver inserted in the RF path, and 3) EOM drivers inserted in both the LO and RF paths.

I could not understand why the measurement limit is so high. Also the case 2 seems too low comsidering the noise level for 1) and 3).

At least we could see clear increase of the noise between the case 1) and 3). Therefore, we can infer the phase noise added by the EOM driver from the measurements.

Note: The additional phase noise could be associated with the original amplitude noise of the oscillator and the amplitude-to-phase conversion by the variable attenuator. This means that the noise could be corellated between two EOM drivers. The true test could be done using a PLL with a quiet VCO. Unfortuantely I don't have a good oscillator sufficient for this measurement.

Dark current of the HQE PD and other PDs were measured.

- The HQE PDs were loaded on the new PD transportation cages (Attachment 1)
The PDs are always shorted by a clean PD plugs. The PD element is still capped with Kapton seals.

- The assignment of the container/slot and the PDs are as follows

- The measurement has been done with KEITHLEY sourcemeter SMU2450.

- The result is shown in Attachment 2. Most of the PDs show the dark current of ~3nA at 15V bias. C1-05 and C1-07 showed higher dark current at high V region. We should avoid using them for the aLIGO purpose. I hope they are still OK at low bias V if there is no noise issue (TBC). You can not read the PD names on the plot for the nominal ones, but that's OK as they are almost equivalent.

- As a comparison, the dark current of a C30655 (serial #10) was measured. Considering a DC current due to an anbient light (although the PD was covered), the dark current of the HQE PD seems double of C30655.

- Taking an advantage of having the setup, I took the same measurement for the Laser Comp. PDs in ATF. I gave the identification as #1 and #2. #1 has full-length legs while #2 has trancated legs. As Zach reported before, they showed significantly high dark current. (Attachment 3)

Transfered for RGA scan

B4 (C1-05) -> F1
C1 (C1-07) -> F2

Dark noise measurement for 6 HQEPDs and 1 C30665. All of these showed sufficiently low dark current noise levels compared with the noise level of the DCPD preamp. The measurement was limited by the input noise (ADC) noise of the FFT analyzer as the line noises were too big.

The measurement has been done with the transimpedance of 1e7. The bandwidth of the measurement was 50kHz.

For the production of the aLIGO PDs, the following transfer of the PDs were carried out
A1-23 Cage A1 -> G1
A1-25 Cage A2 -> G2

The cage A will be baked at 75degC to see if this improves AMU=64 emission.

At the same time, we will put C1-05 (F1) and C1-07 (F2) into the contamination test cavity.

Possible reduction of the QE was observed after air-bake at 75degC.

Yesterday I received Cage G from Bob for intermediate test of the PD performance after air bake but before vacuum bake.
This cage was prepared to be the production pair.

According to the ICS, https://ics-redux.ligo-la.caltech.edu/JIRA/browse/Bake-8047
the PDs were air baked at 75degC for 48 hours.

I took the PDs to my lab to check if there is any issue in terms of the performance.
- Dark current: No change observed
- Dark noise: No noise increase observed
- QE: Probably reduced by ~0.5%.

Here I attached the result of the QE measurement. I have measured the QEs of the baked ones (A1-23 and A1-25) and the reference. Since the reference PD has not been baked, this gives us the measure of the systematic effect. The reference showed the reduction of ~0.1%. Assuming this reduction came from the systematic effect of the measurement system, I observed at least 0.5% QE reduction (A1-23). Note that the previous measurement of 99.8% for A1-25 was too high and dubious. But both A1-23 and A1-25 showed ~0.4% lower QEs.

So I believe the air-baking process reduced the QE.

Another evidence was that now I could clearly see the beam spots on these air-baked-PDs with an IR viewer when the PDs were illuminated with a 1064nm beam. Usually it is difficult to see the spot on the PD. The spot on the reference PD was still dark. So this difference was very obvious. I was afraid that something has been deposited on the surface of the photosensitive element. The surface of the diodes looked still very clean when they were checked with a green LED flash light.

All survived PDs have been measured.

Attachment 1: PDH amp RF part (before the preamp was installed)

Attachment 2: RF-AF transmission

Attachment 3: Attachment 3: LO dependence

Attachment 4: RF amp gain (saturation)

Attachment 5: Input/output noise level

Attachment 6: Attachment 6: Preamp/DCPD out buffer AF circuit

attachment 6: DCPD preamp looks like the opamp is wired for positive feedback?

The polarities indicated in the right circuits were opposite, obviously.

[Rich Koji]

3IFO EOM (before any modification) was tested to measure the impedance of each port.

The impedance plot and the impedance data (triplets of freq, reZ, imZ) were attached to this entry.

[Rich Koji]

The impedances of the individual components from the 3IFO EOM (before modification) were tested.
Each component was modeled by LISO. The LISO model (in PDF and txt) are attached at the end of the entry.

Coils
There are three inductors taken from the EOM unit. They showed the Q ranging from 150~300.
Their impedances are compared with the coil taken from the 9MHz port of the spare EOM (=current LHO EOM).
The inductance of the 8.7MHz inductor indicated higher L but still higher Q.

Todd made a replica of the 45.3MHz coil. He used a silver plated wire and it actually showed highest Q of ~400.

Crystal capacitance
The crystal capacitances were measured by attaching a test rig on the DB15 connector of the crystal housing. The rig was calibrated such that the impedances of the attched components on the rig were measured. They showed somewhat similar feature with parasitic resonances at ~50MHz. Above this frequnecy the capacitance went down (i.e. Abs(Z) went up). This indicates there are stray series LCR in pararrel to the crystal. Not sure what is the cause of this.

The central (24.1MHz) port showed smaller capacitance. This probably means the plates for the central port is smaller. Not sure the actual dimensions of the plates for this unit.
 

Koji, Rich

We took apart the unit removed from the 3rd IFO (Unit serial number aLIGO #3, XTAL 10252004) to see what makes it tick.  Koji has done a fine job of adding the plots of the impedance data to this log book.  Attached are some details of the physical construction showing the capacitor values used in shunt before the coils.

Attached please see my notes summarizing the models for the electrodes and inductors within the 3rd IFO EOM

The circuit models for the 3IFO EOM (before mods) were made using LISO.
Then the modification plan was made to make it a new LLO EOM.

Impedance data, LISO model, Mathematica files are zipped and attached at the end.

[Rich Koji]

The 9MHz port was tuned and the impedance was measured.

[Rich, Dean, Koji]

Stuffed all inductors for the new LLO EOM. As the impedances were sensitive to the positions of the inductors in the housing, they were glued with a glue gun.
Also the lid of the housing significantly change the stray capacitance and lowers the resonant frequency (meaning lowers the Q too), we decided to tune the matching circuit without the lid.

The attached plots show the measured impedances. They all look well tuned and matched. We will prepare and perform the optical measurement at the 40m.

From Cryo Cav setup

Borrowed LB1005 Servo box -> OMC

Dark current measurement for InGaAs QPDs (OSI FCI-InGaAs-Q3000) has been done using Keithley 2450 and Frank's diode test kit. Frank's setup uses various custom instruments which are no longer exist, therefore the kit was used only for switching between the segments.

The diodes were serialized as 81, 82, 83, 84, continuing the numbering for the OMC QPDs. The numbers are engraved at the side and the back of the diode cans.

Overall, the QPDs nominally indicated the usual dark current level of <10nA.
SEG1 of #82 showed a lower voltage of reverse breakdown but this is not a critical level.
#83 showed variations between the segments compared to the uniform characteristics of #81 and #84.

To know any anomaly to the junction capacitance of the QPD segments, the RF impedances were tested with a hand-made impedance measurement.
All segments look almost identical in terms of capacitance.

Measurement setup:
The impedance of a device can be measured, for example, from the complex reflection coefficient (S11). To measure the reflection, a bidirectional coupler was brought from the 40m. Attachments 1 and 2 shows the connection. The quantity A/R shows S11. The network analyzer can convert a raw transfer function to an impedance in Ohm.

Calibration and Measurement limit:
The network analyzer was calibrated with 1) a piece of wire to short the clips 2) 50ohm resistor 3) open clips. Then the setup was tested with these three conditions (again). Attachment 3 shows the result. Because of the impedance variation of the system (mainly from the Pomona clip, I guess), there looks the systematic measurement error of ~1pF or ~25nH. Above 100MHz, the effect of the stray impedance is large such that the measurement is not reliable.

The setup was tested with a 10pF ceramic capacitor and this indicated it is accurate at this level. The setup is sufficient for measuring the diode junction capacitance of 300~500pF.

Impedance of the QPD segments:

Then the impedances of the QPD segments were measured (Attachment 4). The segments showed the identical capacitance of 300~400pF level, except for the variation of the stray inductance at high freq, which we can ignore. Note that there is no bias voltage applied and the nominal capacitance in the datasheet is 225pF at 5V reverse bias. So I can conclude that the QPDs are quite nominal in terms of the junction capacitance.

(Ed: 11/23/2020 The RF components were returned to the 40m)

The dark noise levels of the four Q3000 QPDs were measured with FEMTO DLPCA200 low noise transimpedance amp.

The measurement has been done in the audio frequency band. The amp gain was 10^7 V/A. The reverse bias was set to be 5V and the DC output of the amplifier was ~40mV which corresponds to the dark current of 4nA. It is consistent with the dark current measurement.

The measured floor level of the dark current was below the shot noise level for the DC current of 0.1mA (i.e. 6pA/rtHz).
No anomalous behavior was found with the QPDs.

Note that there is a difference in the level of the power line noise between the QPDs. The large part of the line noises was due to the noise coupling from a soldering iron right next to the measurement setup, although the switch of the iron was off. I've noticed this noise during the measurement sets for QPD #83. Then the iron was disconnected from the AC tap.
 

I see that these measurements are done out to 100 kHz - I guess there is no reason to suspect anything at 55 MHz which is where this QPD will be reading out photocurrent given the low frequency behavior looks fine? The broad feature at ~80 kHz is the usual SR785 feature I guess, IIRC it's got to do with the display scanning rate.

The amplifier BW was 400kHz at the gain of 1e7 V/A. And the max BW is 500kHz even at a lower gain. I have to setup something special to see the RF band dark noise.
With this situation, I stated "the RF dark noise should be characterized by the actual WFS head circuit." in the 40m ELOG.

I consulted with Margot about the cleaning of the optics

• Optics are considered as a clean object. Large dusts can be removed by ionized N2 flow etc.
• Barrel of optics can be wiped with Acetone.
• Optical surfaces are best to be cleaned by First Contact.
• A peek mesh should be embedded in the first contact so that the First Contact sheet can be easily removed.
• When peeling a F.C. sheet from a mirror surface, ionized N2 should be brown for discharging.
• If there are residuals visible on the mirror surface, it should be removed by Acetone. Don't use alchols.
• Use paper lens tissue for wiping as the lint free wipe can be eaten by Acetone.
• In fact, All of the procedure is described in a certain document.
• For a small amount, Margot can provide us a bottle of F.C. and some PEEK meshes.

Details of the Ionized N2 system

• This N2 should have higher purity than 4N (UHP - Ultra High Purity). This means we should use 4N - UHP or 5N - Research Grade.
• The ionized gun used in the clean room at Downs: made by Terra Universal.com
• Flow path: N2 cylinder - Filter - Gun

Floor wiped with a wet wiper (Aug 13, 2012)
Floor wiped with a wet wiper (Aug 15, 2012)
Floor wiped with a wet wiper (Sep 25, 2012)
Air conditioning prefilter replaced (Sep 25, 2012)
Floor wiped with a wet wiper (Oct 01, 2012)
Floor wiped with a wet wiper (Nov 06, 2012) / ATF too
Floor wiped with a wet wiper (Jan 04, 2013)
Floor wiped with a wet wiper (Mar 23, 2013)
Floor wiped with a wet wiper (Apr 17, 2013)
Air conditioning prefilter replaced (Apr 17, 2013)
Floor wiped with a wet wiper (Jun 24, 2013)
Removing Vladimir's mess. Floor swept with a broom (Jun 26, 2013)
Completed removing Vladimir's mess. Floor swept with a wet wiper (Jun 27, 2013)
Air conditioning prefilter replaced (Sep 12, 2013)
Floor wiper head replaced. (Dec 10, 2013)
Floor wiped with a wet wiper (Dec 10, 2013)
Floor wiped with a wet wiper (Apr 1, 2014)

Air conditioning prefilter replaced (Dec 30, 2014)
Air conditioning prefilter replaced (some time in 2015...)

Floor wiped with a wet wiper (Dec 1, 2015)
Floor wiped with a wet wiper (Aug 23, 2016)
Air conditioning prefilter replaced (Aug 8, 2017) = 1 stock remains

Aug 13, 2012 / 0.5um 1000~2000/(0.1 cu ft) / 0.7um   400-600/(0.1 cu ft) by ATF particle counter (MET ONE 227A)

They are counts/(0.1 ft^3)! These numbers should be multiplied by 10 to know the particle "CLASS".

Particle counts

Before the prefilter is installed: 0.5um 1191cnts, 0.7um 346cnts

2:20 prefilter installed
2:25 0.5um 650 / 0.7um 255
3:00 0.5um 578 / 0.7um 99
4:00 0.5um 480 / 0.7um 102
5:00 0.5um 426 / 0.7um 76

They are counts/(0.1 ft^3)! These numbers should be multiplied by 10 to know the particle "CLASS".

1. It turned out that the particle counter MET ONE 227A at ATF shows
(particle count)/(0.1 ft^3)

This means that the numbers I saw previously should be multiplied by 10.
So the nominal class of the room was 5000.

2. As our GT-321s have no diffuser, I borrowed a diffuser from 227A.
The diffuser actually increases the count. We need to buy them.
All the measurments below are performed with the diffuser and calibrated in Count/ft^3.

3. Measured the particle level without the HEPA running.

With diffuser: [cnt/ft^3]

Over Class 10000

4. The two HEPA fans are turned on at the speed "MED".

Basically no particles are detected in the HEPA booth.

With diffuser, inside of the HEPA booth:

The particle level in the room (outside of the HEPA booth) is also improved

With diffuser, outside of the HEPA booth GT-321 #1:
0.3 um 18612
0.5 um   1728

5. The two HEPA fans are turned on at the speed "LOW".

Particle levels are still zero inside.

With diffuser, inside of the HEPA booth, GT-321 #1:
0.3 um 0
0.5 um 0

The particle level in the room (outside of the HEPA booth) is also improved
but the cleaning power for 0.3um seems degraded.

With diffuser, outside of the HEPA booth, GT-321 #1:
0.3 um 34488
0.5 um   1386

Steve and I visited Margot to have a training session for application of First Contact on optics.

- Make "thick" layer of first contact. It becomes thin when it gets dried.

- Apply more FC once a peek sheet is placed on the FC

- Wait for drying (~15min)

- Rip off the FC layer by pulling a peek tab. Make sure the ionized N2 is applied during ripping.

- Margot has a Dark Field Microscope. We checked how the dusts are removed from the surface.
There are many dusts on the mirror even if they are invisible. First Contact actually removes
these dusts very efficiently. Margot told us that even carbonhydrates (like finger prints) can be removed by FC.

The OMC is in the air bake oven now.

Suddenly something dirty emerged in the lab. What is this? It looks like an insulation foam or similar, but is quite degraded and emits a lot of particulates.

This does not belong to the lab. I don't see piping above this area which shows broken insulation or anything. All the pipes in the room are painted white.

The only possibility is that it comes from the hole between the next lab (CRIME Lab). I found that the A.C. today is much stronger and colder than last week. And there is a positive pressure from CRIME Lab. Maybe the foam was pushed out from the hole due to the differential pressure (or any RF cable action).

https://www.airscience.com/purair-flow-laminar-flow-cabinets