`CalorimeterHit` and `SimCalorimeterHit` linked with `RecoMCTruthLinker` sometimes have different cell ID

[AuroraPerego]

While using the RelationCaloHit (podio::Link<edm4hep::CalorimeterHit,edm4hep::SimCalorimeterHit>), I noticed that sometimes a reconstructed calorimeter hit and the associated simulated hit have a different cell ID. I'd not expect hits from different cells to be associated, is this expected?
example (from_elem and to_elem are the cell IDs of the hits):

ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6155, ToIndex=6870, from_elem=22518247251248276, to_elem=18361738493319709588
ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6156, ToIndex=3238, from_elem=18423100386128167828, to_elem=22518247251248276
ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6157, ToIndex=2319, from_elem=34058304684557332, to_elem=18422255948316279700

Steps I followed:

source /cvmfs/sw-nightlies.hsf.org/key4hep/setup.sh
git clone --recurse-submodules https://github.com/key4hep/k4Clue.git
git clone https://github.com/iLCSoft/CLICPerformance                                                                                                                                                               

cd CLICPerformance/clicConfig
ddsim --steeringFile clic_steer.py --compactFile $LCGEO/CLIC/compact/CLIC_o3_v14/CLIC_o3_v14.xml --enableGun --gun.distribution uniform --gun.particle gamma --gun.energy "50*GeV" --gun.multiplicity 10 --outputFile 10gamma_50GeV.root --numberOfEvents 10

cp ../../k4Clue/gaudi_opts/clicRec_e4h_input_gun.py .
k4run clicRec_e4h_input_gun.py --EventDataSvc.input 10gamma_50GeV.root 

I then opened the file my_ouput.root with Python:

import uproot as uproot

# use this to load the tree if some of file.keys() are duplicates ending with different numbers
def load_branch_with_highest_cycle(file_path, branch_name):
    file = uproot.open(file_path)
    all_keys = file.keys()
    matching_keys = [key for key in all_keys if key.startswith(branch_name)]
    if not matching_keys:
        raise ValueError(f"No branch with name '{branch_name}' found in the file.")
    highest_cycle_key = max(matching_keys, key=lambda key: int(key.split(";")[1]))
    branch = file[highest_cycle_key]
    return branch

#load file
filename = '/eos/user/a/aperego/fcc/CLICPerformance/clicConfig/my_output.root'
events = load_branch_with_highest_cycle(filename, "events")

# load links
links = events.arrays(['RelationCaloHit/RelationCaloHit.weight', '_RelationCaloHit_from/_RelationCaloHit_from.index', '_RelationCaloHit_from/_RelationCaloHit_from.collectionID', 
                       '_RelationCaloHit_to/_RelationCaloHit_to.index', '_RelationCaloHit_to/_RelationCaloHit_to.collectionID'])

linkWeight = links['RelationCaloHit/RelationCaloHit.weight']
linkFromIndex = links['_RelationCaloHit_from/_RelationCaloHit_from.index']
linkFromID = links['_RelationCaloHit_from/_RelationCaloHit_from.collectionID']
linkToIndex = links['_RelationCaloHit_to/_RelationCaloHit_to.index']
linkToID = links['_RelationCaloHit_to/_RelationCaloHit_to.collectionID']

# load ECAL / HCAL collections
collections = events.arrays(['ECalPlugCollection/ECalPlugCollection.cellID', 'ECALOther/ECALOther.cellID', 'ECALBarrel/ECALBarrel.cellID', 'ECALEndcap/ECALEndcap.cellID', 
                             'ECalBarrelCollection/ECalBarrelCollection.cellID', 'ECalEndcapCollection/ECalEndcapCollection.cellID', 
                             '_ECalEndcapCollection_contributions/_ECalEndcapCollection_contributions.collectionID',
                             'HCALBarrel/HCALBarrel.cellID', 'HCalBarrelCollection/HCalBarrelCollection.cellID', 'HCALEndcap/HCALEndcap.cellID', 
                             'HCalEndcapCollection/HCalEndcapCollection.cellID', 'HCALOther/HCALOther.cellID', 'HCalRingCollection/HCalRingCollection.cellID'])

simBarrel = collections['ECALBarrel/ECALBarrel.cellID']
simEndcap = collections['ECALEndcap/ECALEndcap.cellID']
recBarrel = collections['ECalBarrelCollection/ECalBarrelCollection.cellID']
recEndcap = collections['ECalEndcapCollection/ECalEndcapCollection.cellID']
ecalOther = collections['ECALOther/ECALOther.cellID']
ecalPlug = collections['ECalPlugCollection/ECalPlugCollection.cellID']

simHCALBarrel = collections['HCALBarrel/HCALBarrel.cellID']
recHCalBarrel = collections['HCalBarrelCollection/HCalBarrelCollection.cellID']
simHCALEndcap = collections['HCALEndcap/HCALEndcap.cellID']
recHCalEndcap = collections['HCalEndcapCollection/HCalEndcapCollection.cellID']
HCALOther = collections['HCALOther/HCALOther.cellID']
HCalRing = collections['HCalRingCollection/HCalRingCollection.cellID']

# map taken with the podio-dump on the input file, the collection hashes may change with a different file
collection_hashes = {
    "0x4b6bf95c": simBarrel, #"ECALBarrel",
    "0x407eb17b": recBarrel, #"ECalBarrelCollection",
    "0x868b64f8": simEndcap, #"ECALEndcap",
    "0xc09a156": recEndcap, #"ECalEndcapCollection",
    "0x1cbea96": ecalOther, #"ECALOther",
    "0xd36e0cf8": ecalPlug, #"ECalPlugCollection",
    "0x76db0c4": simHCALBarrel, #"HCALBarrel",
    "0x6daf3b41": recHCalBarrel, #"HCalBarrelCollection",
    "0x4ca8cf89": simHCALEndcap, #"HCALEndcap",
    "0x50d90aed": recHCalEndcap, #"HCalEndcapCollection",
    "0x3ba1679d": HCALOther, #"HCALOther",
    "0xad7abfca": HCalRing, #"HCalRingCollection",
}

collection_names = {
    "0x4b6bf95c": "ECALBarrel",
    "0x407eb17b": "ECalBarrelCollection",
    "0x868b64f8": "ECALEndcap",
    "0xc09a156":  "ECalEndcapCollection",
    "0x1cbea96":  "ECALOther",
    "0xd36e0cf8": "ECalPlugCollection",
    "0x76db0c4":  "HCALBarrel",
    "0x6daf3b41": "HCalBarrelCollection",
    "0x4ca8cf89": "HCALEndcap",
    "0x50d90aed": "HCalEndcapCollection",
    "0x3ba1679d": "HCALOther",
    "0xad7abfca": "HCalRingCollection",
}

# comparison
for ev in range(len(simBarrel)):
    linkFromID_ev = linkFromID[ev]
    linkFromIndex_ev = linkFromIndex[ev]
    linkToID_ev = linkToID[ev]
    linkToIndex_ev = linkToIndex[ev]
    for FromID, FromIndex, ToID, ToIndex in zip(linkFromID_ev, linkFromIndex_ev, linkToID_ev, linkToIndex_ev):
        from_elem = collection_hashes[hex(FromID)][ev][FromIndex]
        to_elem = collection_hashes[hex(ToID)][ev][ToIndex]
        if not (from_elem == to_elem):
            print(f"{ev=}, from {collection_names[hex(FromID)]}, to {collection_names[hex(ToID)]}, {FromIndex=}, {ToIndex=}, {from_elem=}, {to_elem=}")

output:

ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7912, ToIndex=8290, from_elem=19140057907593236, to_elem=32087902547083284
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7913, ToIndex=7569, from_elem=39125300942471444, to_elem=19140057907593236
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7914, ToIndex=6507, from_elem=18418878059618238996, to_elem=39125300942471444
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7915, ToIndex=5398, from_elem=68961261923075220, to_elem=18418878059618238996
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7916, ToIndex=1257, from_elem=18359205437560062100, to_elem=68961261923075220
ev=0, from ECALEndcap, to ECalEndcapCollection, FromIndex=3623, ToIndex=4060, from_elem=18386227701044740157, to_elem=280680418771005
ev=0, from ECALEndcap, to ECalEndcapCollection, FromIndex=3624, ToIndex=2995, from_elem=18434358234149552189, to_elem=18386227701044740157
ev=0, from HCALBarrel, to HCalBarrelCollection, FromIndex=48, ToIndex=240, from_elem=7036930252865802, to_elem=4222094585888778
ev=0, from HCALBarrel, to HCalBarrelCollection, FromIndex=49, ToIndex=184, from_elem=7318405229576458, to_elem=7036930252865802
ev=0, from HCALBarrel, to HCalBarrelCollection, FromIndex=50, ToIndex=139, from_elem=18441959003400438282, to_elem=18441396049152573962
ev=0, from HCALBarrel, to HCalBarrelCollection, FromIndex=51, ToIndex=105, from_elem=18441396049152049674, to_elem=18441959003400438282
ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6155, ToIndex=6870, from_elem=22518247251248276, to_elem=18361738493319709588
ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6156, ToIndex=3238, from_elem=18423100386128167828, to_elem=22518247251248276
ev=1, from ECALBarrel, to ECalBarrelCollection, FromIndex=6157, ToIndex=2319, from_elem=34058304684557332, to_elem=18422255948316279700
ev=1, from ECALEndcap, to ECalEndcapCollection, FromIndex=4839, ToIndex=5438, from_elem=18441959570348703837, to_elem=18403396265873768509
ev=1, from ECALEndcap, to ECalEndcapCollection, FromIndex=4840, ToIndex=3544, from_elem=18416345394704285757, to_elem=18441959570348703837
ev=1, from ECALEndcap, to ECalEndcapCollection, FromIndex=4841, ToIndex=2333, from_elem=18415499814436339773, to_elem=18416345394704285757
ev=1, from HCALBarrel, to HCalBarrelCollection, FromIndex=42, ToIndex=198, from_elem=18442240517031854986, to_elem=6755438098383370
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=130, ToIndex=452, from_elem=18440833206572810283, to_elem=18441677425345560619
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=131, ToIndex=290, from_elem=18446181183886196811, to_elem=18445618246819774539
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=132, ToIndex=260, from_elem=18445055283979354187, to_elem=18446181183886196811
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=133, ToIndex=109, from_elem=18445899713204453451, to_elem=18445336780432474187
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=134, ToIndex=98, from_elem=18445618229638856779, to_elem=18445899713204453451
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=135, ToIndex=96, from_elem=18445336750367178827, to_elem=18445618229638856779
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=136, ToIndex=94, from_elem=18445336754661621835, to_elem=18445336750367178827
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=137, ToIndex=93, from_elem=18445899704614518859, to_elem=18445336754661621835
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=138, ToIndex=92, from_elem=18445618229638332491, to_elem=18445899704614518859
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=139, ToIndex=89, from_elem=18445336754662146123, to_elem=18445618229638332491
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=140, ToIndex=84, from_elem=18445618251112644683, to_elem=18445618242522710091
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=141, ToIndex=83, from_elem=18445055288273797195, to_elem=18445618251112644683
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=142, ToIndex=81, from_elem=18445336763252605003, to_elem=18445055288273797195
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=143, ToIndex=71, from_elem=18445618233932775499, to_elem=18445336763252605003
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=144, ToIndex=64, from_elem=18445899708909486155, to_elem=18445618233932775499
ev=1, from HCALEndcap, to HCalEndcapCollection, FromIndex=145, ToIndex=60, from_elem=18445899713206026315, to_elem=18445899708909486155
ev=2, from ECALEndcap, to ECalEndcapCollection, FromIndex=6476, ToIndex=7300, from_elem=5628868187127901, to_elem=2814268750692445
ev=2, from ECALEndcap, to ECalEndcapCollection, FromIndex=6477, ToIndex=6747, from_elem=5065918227939421, to_elem=4221476120559709
...

[andresailer]

Hi @AuroraPerego

RelationCaloHit comes from MyDDCaloDigi, and not from the RecoMCTruthLinker, right?
https://github.com/iLCSoft/CLICPerformance/blob/487a80b1a75a0ca276f0d3657d5ab4d7c7c797d5/clicConfig/clicReconstruction.xml#L625

So this has nothing to do with the RecoMCTruthLinker, or does it only happen if you run the RecoMCTruthLinker? Or am I missing something?

I get a bit lost in the script.

[AuroraPerego]

Oh yes, sorry I assumed it was created with the other "links" collections by looking at the Python file.
So the correct place to open it would be iLCSoft/DDMarlinPandora, right?

[jmcarcell]

Ciao Aurora, your file is not public so I can not test on it. However I have files obtained with CLD. If I run the following script:

from podio.reading import get_reader

reader = get_reader('output_REC.edm4hep.root')
events = reader.get('events')
for frame in events:
    coll = frame.get('RelationCaloHit')
    for hit in coll:
        # print(hit.getFrom().getCellID())
        # print(hit.getTo().getCellID())
        assert hit.getFrom().getCellID() == hit.getTo().getCellID()

It never fails. To be completely sure you can run it on your file, but I would trust more this way rather than manually solving the indexes. I don't know what could be wrong in your script, I would have to look in detail.

[dudarboh]

Hi all,

I could reproduce the problem by running @AuroraPerego 's reproducible.
And if I run the code from @jmcarcell for the cross-check, it also shows the inconsistent cellIDs for me.

I uploaded SIM and REC files that reproduce the behaviour for me to the cloud drive.

I couldn't find anything suspicious in the DDCaloDigi.
So my next assumption is that something could be wrong with the LCIO/EDM4HEP conversion.

I checked the LCIO output file after the reconstruction, and I don't see any mismatched cellID0 in the RelationCaloHit collection.

I checked with this code:

#include "lcio.h"
using namespace lcio;

void test(){
    auto* lcReader = LCFactory::getInstance()->createLCReader() ;
    lcReader->open("./CLICPerformance/clicConfig/Output_REC_e4h_input.slcio");

    LCEvent* evt = nullptr;
    //----------- the event loop -----------
    while( (evt = lcReader->readNextEvent()) != 0 ) {
    
        auto* relCol = evt->getCollection("RelationCaloHit");
        
        for(int i=0; i < relCol->getNumberOfElements(); i++){
            auto* relation = static_cast<EVENT::LCRelation*> (relCol->getElementAt(i));
            
            auto recoHitObject = relation->getFrom();
            auto simHitObject = relation->getTo();

            auto* recoHit = static_cast<EVENT::CalorimeterHit*> (recoHitObject);
            auto* simHit = static_cast<EVENT::SimCalorimeterHit*> (simHitObject);

            auto recoCellID0 = recoHit->getCellID0();
            auto recoCellID1 = recoHit->getCellID1();

            auto simCellID0 = simHit->getCellID0();
            auto simCellID1 = simHit->getCellID1();
            if (recoCellID0 != simCellID0){
                std::cout<<"reco/sim: CellID0 "<<recoCellID0<<"/"<<simCellID0<<" eq? "<<(recoCellID0 ==simCellID0);
                std::cout<<"  CellID1 "<<recoCellID1<<"/"<<simCellID1<<" eq? "<<(recoCellID1 ==simCellID1)<<std::endl;
            }
        }
    } 
    lcReader->close();
    delete lcReader;
}

However, I could not check the consistency for the cellID1 as it is not saved in the output file. But DDCaloDigi seems to just copy the values, so at first glance it should be fine.

What I also find interesting is that there seems to be an "off-by-one" error somewhere.
Taking the output of @AuroraPerego as an example, for the consecutive FromIndex in the relation when cellIDs start to be inconsistent, it seems that they are still consistent with a shift of one: from_elem[FromIndex-1] = to_elem[FromIndex].

ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7912, ToIndex=8290, from_elem=19140057907593236, to_elem=32087902547083284
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7913, ToIndex=7569, from_elem=39125300942471444, to_elem=19140057907593236
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7914, ToIndex=6507, from_elem=18418878059618238996, to_elem=39125300942471444
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7915, ToIndex=5398, from_elem=68961261923075220, to_elem=18418878059618238996
ev=0, from ECALBarrel, to ECalBarrelCollection, FromIndex=7916, ToIndex=1257, from_elem=18359205437560062100, to_elem=68961261923075220

What I also don't understand is that the number of simulated hits in the LCIO and EDM4hep outputs is not the same when running the same reconstruction chain. The LCIO output has consistently fewer sim hits. Is that expected?
Maybe I couldn't see those inconsistent cellIDs in LCIO, because they are already stripped somehow?

I am checking with:

podio-dump REC.root > dump_edm4hep.txt
anajob REC.slcio > dump_lcio.txt

Stripped to relevant output:

******************EDM4HEP DUMP EVENT 0 ******************
ECALBarrel              CalorimeterHit                         3758 
ECALEndcap              CalorimeterHit                         6473 
ECALOther               CalorimeterHit                         1002 
HCALBarrel              CalorimeterHit                         24   
HCALEndcap              CalorimeterHit                         95   
HCALOther               CalorimeterHit                         0    
LumiCal_Hits            CalorimeterHit                         15   
MUON                    CalorimeterHit                         0    
ECalBarrelCollection    SimCalorimeterHit                      4114 
ECalEndcapCollection    SimCalorimeterHit                      7172 
ECalPlugCollection      SimCalorimeterHit                      1108 
HCalBarrelCollection    SimCalorimeterHit                      156  
HCalEndcapCollection    SimCalorimeterHit                      281  
HCalRingCollection      SimCalorimeterHit                      5    
LumiCalCollection       SimCalorimeterHit                      19   
BeamCalCollection       SimCalorimeterHit                      0    
YokeBarrelCollection    SimCalorimeterHit                      0    
YokeEndcapCollection    SimCalorimeterHit                      0    

******************LCIO DUMP EVENT 0 ******************
ECALBarrel                    CalorimeterHit                3758
ECALEndcap                    CalorimeterHit                6473
ECALOther                     CalorimeterHit                1002
HCALBarrel                    CalorimeterHit                  24
HCALEndcap                    CalorimeterHit                  95
HCALOther                     CalorimeterHit                   0
LumiCal_Hits                  CalorimeterHit                  15
MUON                          CalorimeterHit                   0
ECalBarrelCollection          SimCalorimeterHit             4050
ECalEndcapCollection          SimCalorimeterHit             7091
ECalPlugCollection            SimCalorimeterHit             1105
HCalBarrelCollection          SimCalorimeterHit               60
HCalEndcapCollection          SimCalorimeterHit              197
HCalRingCollection            SimCalorimeterHit                1
LumiCalCollection             SimCalorimeterHit               15
BeamCalCollection             SimCalorimeterHit                0
YokeBarrelCollection          SimCalorimeterHit                0
YokeEndcapCollection          SimCalorimeterHit                0

[jmcarcell]

I can reproduce this, it looks like something is going wrong in the converter.
@AuroraPerego depending on what you want, the DDCaloDigi processor has been ported to Gaudi in https://github.com/key4hep/k4GaudiPandora, so it's possible to run it without using the converter (and thus not having this issue). See https://github.com/key4hep/k4GaudiPandora/blob/main/k4GaudiPandora/options/runDDCaloDigi.py.

[AuroraPerego]

I can reproduce this, it looks like something is going wrong in the converter. @AuroraPerego depending on what you want, the DDCaloDigi processor has been ported to Gaudi in https://github.com/key4hep/k4GaudiPandora, so it's possible to run it without using the converter (and thus not having this issue). See https://github.com/key4hep/k4GaudiPandora/blob/main/k4GaudiPandora/options/runDDCaloDigi.py.

I'll give it a try, thanks!

[dudarboh]

@AuroraPerego,

The actual troublemaker is algList.append(OverlayFalse) algorithm.

If you look at the source code, it modifies the event and appends some overlay SimHits to the existing collections. That is the point from which EDM4Hep and LCIO SimHits collections start to differ in the number of entries, as the LCIO SimHit collections are not updated after this change.

Then, the digitisation processors run on the LCIO SimHit collections w/o overlay. They create relations based on the LCIO SimHit collections w/o overlay. And in the end, you push LCIO recoHits and relation collections to the edm4hep format, and there the indices simply don't correspond with the edm4hep SimHits collections that have overlay appended.

This happens not only for SimCalorimeterHit collections, but for all collections affected by the overlay algorithm.

The first that comes to my mind to avoid this problem is to save the SimHits after the overlay in separate collections.

OverlayFalseLCIOConv = Lcio2EDM4hepTool("OverlayFalseLCIOConv")
OverlayFalseLCIOConv.convertAll = False
OverlayFalseLCIOConv.collNameMapping = {
  "MCPhysicsParticles": "MCPhysicsParticles",
"InnerTrackerBarrelCollection" : "InnerTrackerBarrelCollectionWithOv",
"InnerTrackerEndcapCollection" : "InnerTrackerEndcapCollectionWithOv",
"OuterTrackerBarrelCollection" : "OuterTrackerBarrelCollectionWithOv",
"OuterTrackerEndcapCollection" : "OuterTrackerEndcapCollectionWithOv",
"VertexBarrelCollection" : "VertexBarrelCollectionWithOv",
"VertexEndcapCollection" : "VertexEndcapCollectionWithOv",
"BeamCalCollection" : "BeamCalCollectionWithOv",
"ECalBarrelCollection" : "ECalBarrelCollectionWithOv",
"ECalEndcapCollection" : "ECalEndcapCollectionWithOv",
"ECalPlugCollection" : "ECalPlugCollectionWithOv",
"HCalBarrelCollection" : "HCalBarrelCollectionWithOv",
"HCalEndcapCollection" : "HCalEndcapCollectionWithOv",
"HCalRingCollection" : "HCalRingCollectionWithOv",
"LumiCalCollection" : "LumiCalCollectionWithOv",
"YokeBarrelCollection" : "YokeBarrelCollectionWithOv",
"YokeEndcapCollection" : "YokeEndcapCollectionWithOv",

  }

Then run all the algorithms down the chain on these new collections and ignore the old ones w/o overlay.
I am not sure if the Lcio2EDM4hep converter allows modification of existing collections... I guess not?

Maybe others could come up with a more elegant solution.

[jmcarcell]

I think what @dudarboh says is correct. I ran disabling all overlays and this issue didn't happen. Only to correct, overlay set to false means there is no overlay, however the cuts that Overlay applies still run (you get less hits in the LCIO dump than you do in the EDM4hep dump). Removing the Overlay from the algorithms means that these cuts don't run. This modification in-place is indeed annoying, and when validating ported algorithms I have to always run without Overlay because of this.

This happens not only for SimCalorimeterHit collections, but for all collections affected by the overlay algorithm.

Does this happen with LCIO input? Probably not since by the time it is converted to EDM4hep everything is consistent. So only collections that are affected by the Overlay algorithm and are read from EDM4hep.

@andresailer maybe this should be moved to https://github.com/key4hep/k4EDM4hep2LcioConv since it is an issue in the converter