DC IR VS AC 1kH IR measurements

Announcement - Help us fight the BOTS! Please report all spam including stuff in your inbox!
Joined
Mar 7, 2018
Messages
1,051
Looking at the spec sheet for the INA260 the accuracy is going to be an interesting problem because it's designed to work with relatively stable loading. I was hoping that the trigger mode would allow for an external trigger input (rising or falling edge of the PWM) but the trigger pin is driven from internal criteria only. Even thought the trigger pin could be used to create the PWM by a bit of hokey pokey, yes not in a manner that would make the results easy to interpret, deep rabbit hole. The Fig.13. Frequency Response plot is relatively trouble free upto about 200Hz.

The ADC does not seem to specify a sample and hold type ADC so switching transitions of the load will cause more detectable singular sample errors. This critically also limits the upper load frequency and may require a formula to convert the INA260 values to the "actual" values. This is because if the INA is sampling at say 1mS and the load is 1kHz (50/50) the load is only on for 0.5mS and the measured ADC value will be lower than the actual value by some function as to how the ADC works (e.g. successive approximation). With a rising edge the ADC may chase the measurement to the top but on a falling transition the ADC would stop at some random point and this random value would occur as a function of frequency. As the frequency rises the measured value will decrease more and more below the actual value on sample averaging approach.

I'm wondering if triggered sampling mode (write to register to trigger conversion of one sample only - no averaging) on the INA260 can then remove over 95% of detectable current measurement errors (switch off transitions) by filering with a maximum expected allowable deviation.

The ADS1115 does not have an extrernal trigger either, unfortunately. Searching web for parts...

So... low frequency testing may be better driven from a pin rather than the servo controller PWM output, but the frequency limit will be quite low. Higher frequency will have to average a lot more samples and also be limited below 1kHz without further interpretation of measuements.
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
CC

Got it.
I was thinking the same triggering the MOSFET from the ESP32 initially and also sticking to lowf for now.

At least we can get the prototype functional initially, working out the details as they come along.
It's going to be a challenge as it is, putting something together with cheap off the shelf parts and experimental code for now.
Looking into the ledc channels for proper PWM and interrupts to accumulate data for a specific amount of time? Maybe?

Code:
sei(); //Enables interrupts
delay (ms); //Wait for data (ms) determined by encoder position or some other means
cli(); //Disable interrupts
Also it looks like the ESP32 may have a problem with anything below 15Hz on the ledc channels? Don't know yet.
At first I think I'm just going to manually set up a 10s on/off and see what reaction I getfrom the sensors.
Pull out my math books, which I swore I wouldn't do, and see if the calculations will be right.
That will be step 1.
After that we can go to Step 2 and so on.

Wolf
 

gauss163

Member
Joined
Jun 28, 2020
Messages
278
completelycharged said:
Found the sheet, starting to play.... [...]

Iteresting falling off the cliff


image_rhcfuy.jpg

It'd be interesting to see that with theDC IR plotted too. Below are results from a study of 800 cells that show analogous results. Note how the capacity uncertainty increases as the cells age, e.g. 130m?could be 93-103%, but 160m?could be 5-45% capacity.Presumably that's due to the growth of the non-ohmic resistance - which cannot be seen by the 1kHz AC test. But they didn't do any DC IR tests so we can't be sure.


image_otbhqx.jpg
 
Joined
Mar 7, 2018
Messages
1,051
From that plot it looks like the charged vs discharged Ohm difference might widen with age/capacity as well. Individual cell data through a full cycle will get quite interesting.


The two discharge charts in the paper also show the degredation characteristics a bit better when shown together and the interesting 70-80% point of accelerated degredation.

image_qubosn.jpg


Full cycle testing with enough random sample cells (from typical recycling) would allow for the voltage discharge curves to be derived without having to do 800 cycles on a single cell.... as long as the cells are correctly identified and not mixed up.


Within the bode diagram, thier test data also shows up a slight variation/pattern in the response around 100Hz ???


image_vicgrs.jpg


Would seem a bit strange and carless to have 50/60Hz harminics in the test setup if it is just the supply, otherwise could this small deviation actually be showing something very particular if the resolution was amplified ?


Compare Fig 9 and Fig 10 around the 100Hz point... phase angle indicates less capacitive impedance but Z increasing, so.... non-ohmic internal characteristic ????
 
Joined
Mar 7, 2018
Messages
1,051
Looking at the cell test data, there appears to be a bias between testers or just coincidence that a set of cells were tested on some with some testers.

XTAR reports lower capacity and higher IR than Foxnovo for example. <50mA bias and >5mOhm bias for XTAR.

Messing about with plot for Resting volts X, mAh Y and bubble as mOhm above sample minimum (rebased to normalise)



image_efmlyt.jpg


XTAR top

Based on the cell numbers it appears random..
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
CC
If you are talking about the LGGBM261865 sheet....
Oh yea definite bias.

I noticed that after about 200 cell its seems the slots of each "testers" mAh results where consistently higher/lower.
Look at the OPUS 1 slot 4 it sticks out with always having a higher mAh result. the SKYRC cells 2 and 3 always just a bit higher.
Slot 2 on the XTAR always a little higherand so on.
Also tester finish voltage on the OPUSES (or is it OPI,orOPUSI : :p )
is always a bit lower on OPUS 3 1-4.
IR was not checked with the testers but obviously with the RC3563.

On another note I finally got a very stable V and accuratereading on the ESP32 with the ADC1115after playing with divider resistor values and formulas.
On to the next step Amp readings.
On the prototype board with cobbled together code though I was able to get Voc , Load V and I and did a quick calculation for DC IR with a 10 second load. Preliminary results but encouraging for the first step.

Wolf
 

gauss163

Member
Joined
Jun 28, 2020
Messages
278
Below is a graph that may help to make more concrete some of the things I discussed above. It shows a typical voltage response when the current is instantaneously changed from a higher to lower value (e.g. at the end of charge, when it changes from the termination current to zero). Below I briefly explain the various resistance components that contribute to the voltage decay and give some links for further reading.


image_hxfmnb.jpg


R[size=small][size=small]O =[size=small]O[/size][/size][/size]hmic resistance.The instantaneous voltage drop?V1 is caused by RO, thepurely ohmic component of theresistance (which corresponds closely to the resistance measured by AC 1kHz meters). This is the sum of the resistance from the cell's componentsthat behave like a resistor, i.e. are governed byOhm's law:[size=small][size=small]?[/size][/size]V = R [size=small][size=small]?[/size][/size]I,such asinternal wires(tabs) and electrolyte ionic resistance.

[size=small][size=small]R[/size][/size][size=small][size=small][size=small][size=small][size=small]CT =C[/size][/size][/size][/size][/size]harge Transfer resistance accounts for the voltage drop[size=small][size=small]?[/size][/size]V2typically occurring in the first few seconds, dueto charge transfer at the electrolyte/electrode interface (as well as double layer capacitance).

[size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small]R[/size][/size][/size][/size][/size][/size][/size][/size][/size][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small][size=small]p = p[/size][/size][/size][/size][/size][/size][/size][/size][/size][/size][/size][/size][/size][/size]olarisation or diffusion resistance corresponds to the shallow almost-linear voltage drop off that occursafter a few seconds, caused by ionic diffusion. This is thethe main component of voltage drop observed when removing a cell from a charger, as the cellslowly asymptotically creepsto itssteady-state resting voltage (it mayalso includesome ofthe drop from theprior fastertimescale components if you measure it just before removing from the charger, or very quickly thereafter).

DC IR measurements typically include all of these components (but only an initial part of the hours-longRp voltage drop),depending on the length of the test pulse (and its amplitude). Taking all of this information into account may help to yield a better idea of how the cell's voltage drops under longer timescale DC loads - infothat is much harder to glean from the short timescale 1kHz AC tests. This may in turn lead to moreaccurate measures of cell health.

There is a nice introduction to this and related matters in Barai et al. A study of the infuence of measurement timescale on internal resistance characterisation methodologies for lithium-ion cells,2018. It also includes helpfullinks to prior related work, see esp. the paperSchweiger et al. Comparison of Several Methods for Determining the Internal Resistance of Lithium Ion Cells, which reviews and compares many of the common methods of testing IR = internal resistance.
 

Bubba

Member
Joined
May 9, 2018
Messages
300
I found this quote interesting from https://www.nature.com/articles/s41598-017-18424-5.pdf
from gauss163. Nice read thanks.


"Depending on cell type and experimental setup (e.g. cable assembly), the 1 kHz resistance measurement can
be in the inductive or conductive region. Typically, for cells with large capacities (e.g. a 40 Ah pouch cell) the
resistance at 1 kHz is dominated by inductive behaviour. On the other hand, cells with relatively lower capacities
(e.g. a 3Ah 18650 cell) can have a 1 kHz resistance close to the point where Im(Z(?)) = 0 29, and thus give reliable
and repeatable measurements."

So for single 18650 cells the 1kHz method is ok, but for larger cells or comparing packs it doesn't work well ... Is this conclusion correct?
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
Here is some initial data of a new US18650VT4 cell with 19.185 mΩ AC IR.

Resistor was 4Ω and time was 12 seconds

All preliminary. I'm going to go with 1 or 2 more decimal points on the current measurement for better resolution.

Here is the chart.

Wolf

image_cjdgmr.jpg
 
Last edited:

Redpacket

Member
Joined
Feb 28, 2018
Messages
1,288
For me, there's two angles to this:
1) determining good/bad cells early with an easy test, eg the usual impedance meter 1kHz tests
2) what our packs are doing while in-use.
From the pdf study a few posts up from Bubba, I'm most interested to see the results around our "in-service" use eg figure 4 b) or c) where they show a cell already under some DC load & the load is increased. Or charging the same sort of step.
I don't see much value in charge to discharge steps or in "resting" to load/charge steps.
Also, like mentioned earlier, understanding what happens with the 50/60Hz half-sinewaves out inverters typically draw would be good to figure out!
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
Well the DC IR experiment continues. The prototype is functional and giving good results.
With a 1Ω resistor the numbers look like this on a Sony US18650VT4

RC3563 1kHz AC IR 19.247mΩ 4.209 V

The prototype DC IR Tester

V Open circuit 4.208
V Loaded circuit 4.043
mV Voltage Drop 0.165
mA Circuit draw 3422.50
mΩ DC IR 48.31861


Stay tuned Wolf

image_lpzatr.jpg
 
Last edited:

Redpacket

Member
Joined
Feb 28, 2018
Messages
1,288
So I did a few simple tests on an LGABC41865 cell from a laptop pack.
I did DC load steps.
I don't have a nice test gear as Wolf & some of you guys & it wasn't an automated test but here goes:
Meter A = Protech 506 Multimeter
Meter B = YR1030+ 4wire 1kHz resistance meter
Resistors, 2x 47ohm/5W wire wound measured with meter A = 47.0, 47.2
Resistors, 2x 6.8 ohm/5W wire wound measured with meter B = 6.74, 6.79

Temp =~18 degC
The cell had approx 1cm of typical spot welds straps attached each end which likely added a few m.ohms
Cell volts resting before test: 3.926V
Cell measured approx 50.1 m.ohms with B at 1kHz before tests
Allowed approx 5 sec to stabilize at each load point
Resistors re-checked after tests, no significant change eg due to heating, etc

Test 1:
Load with 1x 47 ohm resistor, measured volts with A, = 3.917V = 83.34mA
Add 2nd 47 ohm resistor, measure volts with A, =3.910V = 166.03mA
So Cell Resistance = delta V over delta A = 7mV/82.69mA = approx 84.6 m.ohms
Cell at rest again recovered to approx 3.924V ~30 sec after test

Test 2: (approx 5 mins after test 1)
Load with 1x 6.8 ohm resistor, measured volts with A, = 3.856V = 570.4mA
Add 2nd 6.8 ohm resistor, measure volts with A, =3.807V = 1.1263A
So Cell Resistance = delta V over delta A = 49mV/555.9mA = approx 88.1 m.ohms
Cell at rest again recovered to approx 3.920V ~30 sec after test

I repeated the tests on the same cell & got similar numbers
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
All right here is where we are so far.
I have managed to cobble together some code for this DCIR tester.
Here was my theory.
Build a 4wire voltage tester with an Amp sensor. Record the V open, V drop, mA draw and calculate the DC IR
as in (V open - V drop) / ma = R
I built this tester using an ESP32 with an ADS1115 V sensor and a INA260 Amp sensor. A 4 wire cell holding fixture and a 1Ω 100W resistor.
The Function Is as follows:
A charged cell is inserted into the fixture and the ESP32 is in a while loop waiting for a push button to exit it and run the main loop.
The main loop measures the V open saves that andengages the MOSFET for 1 second while the 2 sensors record V drop and mA.
All gets spit out in real time to excel using PLX-DAQ.
Once the button is released the while loop resumes.
I have tested several cells and the results are posted below.
The trace of the charts goes from green stripe to green stripe north and south of the green stripes is noise.
For some reason the MOSFET engages for a split second when the button is pushed so the first readings (north of the green line)are right but wrong. :huh:
I have no idea why it does this. Maybe someone can give me a pointer there.
The averages are calculated between the white lines including the blues with results to the right of the charts. Also included is the cell model AC IR and V measurements prior to inserting the cell.
I have run the same cells several times and the results are almost always identical within a couple of mΩs.
One interesting thing is the traces are somewhat different for each cell but same part numbers seem to be similar.
We all understand that DC IR is going to be different and higher and it appears that cells I measured with AC IR and claim in my cheat sheet to be marginal certainly have a high DC IR.
This is the best I have come up with in a quick easy DC IR tester that is adjustable in the code to pretty much anything you want as far as time and measurement parameters is concerned. I'm sure there is much that can be improved with the sketch so if anyone wants to help I will gladly share.
So much more to study and learn..............
Oh yea that lonely led on the left is lit to tell me the while loop is running and the system is ready for a test.
OK pictures as promised.
The Tester:

image_forcka.jpg

The Charts:

image_vpcuce.jpg

image_ckqqjv.jpg

image_mkfodh.jpg

image_tmyxkf.jpg

image_bfqmds.jpg

image_pjhtmp.jpg

image_ycqbui.jpg

image_awvwgi.jpg

image_mjnawu.jpg



Wolf
 
Last edited:

gauss163

Member
Joined
Jun 28, 2020
Messages
278
Glad to see you are making progress. The wide variation between the AC IR and DC IR gives us something to ponder. Is it correlated to capacity? (please list the capacity too, and the discharge rate and termination).

Did you test any of the prior mentioned cells whereAC IR was ambiguous, i.e. some cells with close AC IR had both good and bad capacity? Likely a DC IR test would help to further discern the bad guys.
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
gauss163 said:
Glad to see you are making progress. The wide variation between the AC IR and DC IR gives us something to ponder. Is it correlated to capacity? (please list the capacity too, and the discharge rate and termination).

Did you test any of the prior mentioned cells whereAC IR was ambiguous, i.e. some cells with close AC IR had both good and bad capacity? Likely a DC IR test would help to further discern the bad guys.

Progress yes thank you I'm doing the best I can with my limited knowledge.
The cells tested were a random number of cells that I have just sitting around for just such testing purposes.
At this point it was just to show how the tester works and what information can be gleaned from its results.
Worth mentioning is,that the placement of eithera 4? or 1? resistor as the load value has very little influence on the outcome as initially observed.
I would have both of them hooked to the board with separate MOSFETs but my coding skills are not there yet so Ihave decided to stick with the 1?.
It is easily changed manually if necessary.
Are the results telling us anything? IDK yet. I do know they are consistent for each cell. I have tested several cell multiple times and the results are always very close.
I will redo the tests and post those charts also.
When the time permits I have brand new cells and a bunch of tool and medical packs to break apart and start a sheet on this testing procedure
taking each individual cell from pack liberation to potential healthy cell.

@gauss163
"Is it correlated to capacity" I don't believe so and yes I intend to gather all this data with cells I have never tested.
Asfar as close AC IR between same manufacturer part numbers I have never had a bad cell with proper AC IR,i.e. if the AC IR was 45m? on 1 cell and 44m? on another both cells would usually be within ~100mAh of each other.
Also my AC IR cheat sheet has never let me down. If a cell is at the margin of the upper limit it may be a crap shoot but if the cell is well within the range you are 99% guaranteed to get a good cell.
As I test these cells coming out of these packs it will be interesting.
The new sheet will be very thorough recording all the pertinent measurements for these tests.
UnfortunatelyI am only 1 person and do not have a lab other than what you see.
Also I still work to make a living (bummer) as I would love to do this all day long.
Additionally I still need to finish my second 14s80p pack........

Stay tuned
Wolf
 

gauss163

Member
Joined
Jun 28, 2020
Messages
278
daromer said:
Overmind said:

Because its true on its normal span of cycles/lifecycle. IR do not change much during a normal life. [...]

As is clear fromprior posts here,AC IR does in fact change during normal cycling, but the change is much smaller than for DC IR, because DC includes additional non-ohmic components,e.g. see post #43, and post #12 (and it isalso implicit in wolf's data). That's why both IR measures are used to test battery health.

As I warned before: info atBatteryUniversityis highly unreliable - don't trust it.
 

Araknid

Member
Joined
Jun 1, 2020
Messages
50
Could do with this fancy looking 4 wire cell holder too ;)

Makes loading cells into the tester soo smooth :) terminals pivot around a gear and engage the 4wire pickups on the battery ends.. ahhh


image_halynh.jpg
 

Wolf

Active member
Joined
Sep 25, 2018
Messages
1,563
Here is an update on DC IR VS AC IR.
The tester I built seems to be working quite well. I have tested ~100 cells so far and have all this info recorded on this sheet.
https://drive.google.com/file/d/1w1s3S6Zsxx-4ejtWsY6R9swlQFXtPLn0/view?usp=sharing

Synopsis of the results. (?) = average of all cells per pack

First pack was a medical pack with Moli ICR-18650H (2100mAh) cells.
The AC IR was so high I would have never even thought about using these cells but for sake of the experiment I did. My preferred IR for this cell was ?50m? these came through at ? 78m?.
The Voltage was decent at ? 3.7V DC IR was ? 334m?. Woah that was high is my tester working right? Tested the cells and their SOH results, after C/D/C was ? 28%. After charge AC IR actually went up ? 3.6% and DC IR went slightly down ? 6.8%

image_btaxxj.jpg


Second pack was another medical pack with Sanyo NCR18650BF (3200mAh) cells.
AC IR was reasonable at ? 41m? unfortunately I was unable to find a factory spec on AC IR so I assumed it was OK. Voltage was a little low on the cells ?2.3V. Checked DC IR at that level and found it to be ?240m?. Recovery Charge at 100mA to 150mA to 3V and then full charge. SOH results, after C/D/C ? 100%. AC IR dropped a bit ? 4.2% and DC IR dropped dramatically which is to be expected ? 64%.

image_pgbnve.jpg


Third pack was a Hoover Commercial 40V pack with Samsung INR18650-30Q (3000mAh) cells.
AC IR was ? 12.6m?. Voltage was ? 3.8V except for 4 cells which were at ? 0.39V. Probably the cause of the demise of the pack. DC IR was not done (on the 4 low V cells) as there was no voltage to drop. AC IR was higher than the others but still below factory spec (?18m?) at ? 13.3m?. Nevertheless recovery charge at 50mA to 3V and they were tested with the others. DC IR on the other cells was ? 26.3m?. Wow thats low. So I started looking at spec sheets and lo and behold I found a spec sheet for the Samsung INR18650-30Q listing AC IR and DC IR. To my surprise the DC IR that I measured matched the manufactures spec at ?30m?. I was feeling good about my tester.
SOH results, after C/D/C, was ? 100% including the 4 low V cells. After charge AC IR dropped slightly ? 2.7% DC IR dropped slightly also ? 3.1%.

image_odojbu.jpg

image_ypjkoa.jpg


Fourth pack was another medical pack with Moli ICR-18650K cells. Manufactures AC IR states ?80m? but in my experience and my cheat sheet ?70m? is preferred and a max of 79m? although that max may need to be revised in my cheat sheet. AC IR came in at ? 85.5m? Voltage was a respectable ? 3.8V and DC IR was ? 303m?. SOH after C/D/C was 42%. After charge AC IR decreased ? 1% and DC IR decreased ? 8.5%.

image_gzmywa.jpg


Fifth pack was an EGO power mower pack with Sanyo UR18650RX (1950mAh) cells.
Manufactures AC IR <25m?. AC IR was ? 12.1m? Voltage respectable ? 3.7V and DC IR was ? 27.6m?. SOH after C/D/C was 103% (skewed because all where tested with OPUS). AC IR dropped negligible by ? 0.02%. DC IR dropped by ? 9.35%

image_dvpxlh.jpg


Sixth set of cells did not come from a pack but some cells I have collected for a 12V battery.
Picked 12 random cells out of the 40+ I have.
The set was comprised of 12 Samsung INR18650-20Q (2000mAh) cells. Manufactures AC IR spec was ?18m?. My cheat sheet was at ?25m? and up to 35m? with a ? Not enough data.
AC IR was ? 25m? Voltage was ? 4.08V and DC IR was ? 58.9m?. SOH after C/D/C was ? 97%
AC IR dropped ? 1.5% and DC IR dropped ? 8.1%

image_fcpsoh.jpg


Seventh pack HP supplementary 12 cell battery I have had for a while. (I have 20 more of them)
But knowing they were fitted with 2200mAh cells I didnt take them apart yet. Nevertheless now was the time for the experiment.
The pack had 12 Samsung ICR18650-22F (2200mAh) cells with an AC IR of ? 58.4m? 2.48V and a staggering DC IR of ? 291.7m?. My cheat sheet claimed preferred status of ?65m? and up to 75m?. SOH after C/D/C was ? 97% (cheat sheet was right as far as AC IR was concerned) some of the slightly higher V cells had lower DC IR. AC IR dropped by ? 1% and DC IR dropped by ? 61.6%

image_nonikd.jpg

And just for kicks here is a screen capture of the DC IR results and how they are calculated. This happens to be cell #0099 1st DC IR Test after C/D/C.

image_usssqd.jpg



Please feel free to download or view this sheet for further study if interested.
File is attached
Several observations and a conclusion.
To demonstrate the validity and consistency of my DC IR tester, the after C/D/C DC IR test was done twice on the same cell. The differences between the 1st and 2nd test are marginal at the least.
So as far as consistent results yes the tester is consistent. As is the RC3563 AC IR tester.
As far as AC IR is concerned it should be the first touch of the cell to determine the feasibility and the potential health of the cell. The advantage of AC IR is that the cell can be tested at a lower voltage and still give accurate results. The same cannot be said for DC IR the cell needs to be charged to at least 4.xx V for a somewhat reasonable result.
Commercial testers IR measurements (OPUS SKYRC etc.) are haphazard at best. At times they are close other times they are in left field.
DC IR is a neat thing to investigate and for sure can indicate a bad cell but the cell has to be charged to be able to determine that. Well if you dont check anything on the cell and C/D/C it and come up with a SOH of 23% I dont need to be a rocket scientist to determine this cell is N/G.
On the other hand take the cell before a C/D/C cycle, get an AC IR reading of m?s that is out of spec then there is no need to go any further with the cell wasting your time on an underperforming cell.
Additionally DC IR at the initial cells harvest state is just not possible most of the time as the SOC of the cells we harvest mostly are at storage levels or below.


So my conclusion follows this presentation done in February, 13, 2020
https://fhi.nl/app/uploads/sites/74/2020/02/Batenburg-Mechatronica.pdf

Conclusions
Low frequency AC-IR is a valid alternative for DC-IR measurement
AC-IR reduces measurement time dramatically
Possible to measure resistance of 1m? or less accurately
4-terminal pair test leads required to measure low resistance with AC-IR

I could not have said it better myself.

Additional reading: https://www.electronicdesign.com/te...e/21128843/measuring-dcir-of-lithiumion-cells

Any questions please feel free to ask and comment.
Wolf
 

Attachments

  • AC DC IR Comparison.xlsm
    928.6 KB · Views: 85

Redpacket

Member
Joined
Feb 28, 2018
Messages
1,288
That's some awesome work there.
I guessing the DC IR results also followed the discharge current rating for the cell?
Ie it would confirm high current cells would have a lower DC IR than say your average laptop cell?
A 1ohm load would be quite a high load for some cells right?
 
Top