Digital Modeling of Amps: Kemper vs Fractal

The digital modeling world has advanced tremendously in the past few years.  Two products have emerged at the top: the Kemper profiling amplifier (KPA), and the Fractal Audio Axe-FX. [Note at the time of this posting, the Line6 Helix has been pre-introduced but is not yet available for review. It does look promising and will be reviewed when available.] Both the KPA and Axe-FX are exceptional in what they do, but for me, one stands out and offers some unique capabilities.

Here’s some background. I have owned an Axe-FX Ultra since 2008 and a Kemper from 2013, and have been happy with both. [BTW, I have listened and played with the latest Axe-FX II also (but not the Quantum FW). It did not impress me enough to spend the cash to upgrade from either the Ultra or the Kemper.]

For the Fractal, from a hardware perspective, the only minor issues have been the flimsy front-panel buttons and the noisy fan. Overall, it has been pretty good and it gets great sounds. However, I don’t use it as much these days. The main cons of this unit are

• Work flow is not the easiest — going through all the menus and looking at the small LCD screen. You really need the Axe-Edit software to edit the thing (old version since Fractal doesn’t support the Standard or Ultra anymore unfortunately)
• Constant change in firmware updates which change patch sounds. This is a plus as it fixes issues and adds more features, but a con if it adds those features and amp models you don’t care about. This doesn’t apply to the Ultra since no more development work is done for discontinued models.
• No more new firmware or update support for Standard and Ultra models
• Way too many esoteric options
• Tendency to favor more expensive models and phase out support of older models early

The Axe-FX II improved on things by making more models available, better modeling, more effects, headphone out and a USB audio interface.  Later models improved memory, IRs, etc. The latest model is the Axe-FX II XL+.

This article will discuss more in detail a comparison in the methodology in how the sounds are achieved (as I have gleaned by reading the forums and articles). It isn’t going to do a feature by feature comparison or cover detail on the controls, etc.

General

In general, people like the Axe-FX II better for its effects. It uses 24b 48Khz processing as opposed to the KPA’s 24b 44.1kHz.  The sound quality here can be debated, but I cannot hear a real difference, especially live or in a mix. The Axe-FX II has far more effects and parallel routing options.  It also has a PC editor, and its standard IRs are 2048b in length (~44ms, longer with the proprietary Ultra-res IRs. Some people don’t think the added length makes a difference in capturing cab responses (me included), but that can be debated).  Some main users of the effects unit are Alex Lifeson of Rush, John Petrucci of Dream Theater. Metallica have also been using it for all the live sounds. This is a strong endorsement. Axe-FX II price is about $2200 USD, higher for the models with more memory like the II XL+. You can only really buy the higher end models which run closer to $2.6K USD. Price has dropped recently to $2250 — most likely due to competition from Kemper and Line6. 

For the KPA, people like the workflow and tone, and how it is setup like an amp, easy to use buttons and knobs to edit your sounds as well as the ability to profile your own gear.   For pure horsepower, the Axe-FX II is more powerful (it has 2 ADI TigerSHARC DSPs), but the KPA holds its own in tone. The consensus is that it is simpler and faster to get a tone out of it. It does not have a USB audio path. A large number of European metal bands use it live, and Mike Wagener (very famous engineer/producer) uses it in the studio and on albums. Price is $2000 USD in 2015.

Though I own the Axe-FX Ultra, the KPA appeals to me more than the Axe-FX. Read more to see why.

Philosophy

Fractal

Fractal’s underlying technology uses sophisticated behavioural models of electronic components to model amps and effects. The electronic schematic is the key essential component here, as that determines how the circuits are put together. It is doing realtime circuit simulation. It isn’t likely to be doing SPICE-like simulation (used in modeling analog circuits based on the underlying solid-state physics of the devices) since that isn’t possible with complicated circuits in realtime yet — even on Intel x86 i7 hardware (2013).   Most likely, each component is modeled as an abstract behavioural model, with variable parameters to mimic real world behaviour.  Improvements in the Axe-FX are usually due to improvements in the component models. This has been stated in the past by the founder, Cliff Chase. This approach is also similar to what Scuffham does with his SW modeling product, S-Gear.

Kemper

Kemper approaches it in a very different manner which no other modeler up to now has explored.  This is what makes it truly innovative — more so than what Fractal (or anybody else for that matter) does. The Kemper approach is also patented.  Christophe Kemper is the founder, and his background is in synthesizers — as he developed the Virus synthesizer line. His approach is to model real amplifiers as a complete blackbox distortion generator. Kemper sends signals to the signal chain of a real amp rig (could be pedal+amp+cab+mic+pre) and analyzes the output to understand how it generates the tone.  My guess is that he is analyzing the frequency response, impulse response, and specific distortion characteristics (dynamics, overtone generation, intermodulation distortions, etc.) and extracting certain parameters which he then uses in a global distortion model which recreates the sound of the amp rig.  This is quite novel.  It is a high-level approach to modeling as opposed to Fractal’s low-level schematic-based component modeling. I am speculating that this idea works because most distortions coming from devices for guitars (whether it be solid-state, tube, etc.) have certain common characteristics. Once you understand the underlying behaviour, you can create a distortion generator based on it.

Sound Options

The biggest advantage of the KPA is that it can model almost any typical amp that we know that creates a pleasing distortion. If you have an old amp which has a specific sound, a circuit schematic may not recreate it properly. This is because the characteristics of the components could be very different than what is in the schematic.  You would have to take it apart and analyze it if using the Fractal method, and change the firmware to include that model. With the KPA — anybody can profile their amp — including the entire signal chain such as pedals, cabs, mics, and mic pres — and capture that sound.  Fractal uses Impulse Responses (IRs) to capture the cabs/mics/pres, but you need to get those from Fractal and others unless you have all the equipment. KPA allows you to do that yourself on your own amps as the profiling method is part of the product.  It also allows you to separate out the cabs/mics from the basic sound.  In addition, once you profile an amp, the profile is really a bunch of parameters that control the distortion algorithm in the KPA as well as the IR of the cab. Thus, it is very small in size and easily shared with others.  This is a huge advantage.

Some Axe-FX proponents would say that you really need to tweak any base sound to get a sound you like. I think this is exaggerated.  Almost everyone gravitates to an existing amp sound as a base tone they use: be it Fender, Marshall, Vox, or Mesa Boogie. If new tones were really the rage, you’d see many new amps modeled inside the Axe-FX that don’t exist. But yet the majority of amps modeled in the Axe-FX all exist in reality and it’s what everybody uses!  The KPA allows you to model that great sound you might hear from an amp at a specific setting. Most guitar players choose an amp they like and tweak to get a sound and setting. Then they rarely change those settings. So the KPA fits well here. You just profile that sound setting.  There’s still the ability to adjust gain and eq afterwards, so this is a very cool concept.  Kemper’s method can analyze how the gain structure works and he can adjust it down. This means you can profile a high-gain sound, and then later in the KPA, adjust it down. Very unique!

Another main advantage of the KPA is that it can model all kinds of amps and signal chains without knowing the schematics. Even if you did know the schematics, the device values and behaviour could be significantly different than what is in the Fractal component model, So you could really never model it correctly and capture the sound. This is why I believe Fractal uses so many parameters in his model — so he can account for all the variations. But then this makes it way too unwieldy to tweak and adjust. One look at all the parameters for any given amp model can attest to this (it is a tweaker’s dream).  Fractal can only model amps that exist in the firmware. Don’t see an amp in the Axe-FX that you like? You’re out of luck. The KPA on the other hand can model countless amps (and pedals) with its profiles — which are very small in size and can be easily exchanged. In 2015, there are 6000+ user-created profiles on the Rig Exchange (all free), with many more commercially available.  Including free and commercial, there are probably 10000+ profiles available.

Workflow

The KPA also is built so that the workflow is like an amp. You don’t have flexibility in routing like in the Axe-FX. But how many real-world setups (live or recorded) have parallel multi-amps going and 4 streams of it in stereo? In fact, most great amp sounds may have multiple sound sources, but ultimately are mixed to one track (this is shown in many interviews with leading recording engineers).  Also, in reality, you can place the effects and things in almost any order, but we already know from experience that this is rarely done, and optimal effect orders are known.  Thus, though the KPA is more restrictive, it fits 99% of what most people do. For the 1% who love to tweak, make crazy sounds and do weird or complex routing — the Axe-FX is probably for you. But the general workflow in the KPA: input → stomps → stack → effects → output is really applicable to 99.9% of all guitar rigs and sounds.

Cabinets and Impulse Responses

The KPA profile files are so small in size (~4KB) that sharing them is inevitable. There are so many great free and commercial profiles that this is really what makes the KPA shine. New and interesting sounds are constantly available. This is not possible with the Axe-Fx unless a new firmware update is done, and often times this perturbs previous patch tones.  You can get new cabinet IRs, but sonically it only changes an existing amp sound — not a completely new amp sound (though the IR contributes a lot to the sound).  Fractal is aggressively pursuing this area as it allows them to make money by continuously offering new IRs and sounds — without firmware updates. In a sense, he is copying the business model of Kemper, who really have a great platform that leverages many user-created profiles.  

Another big advantage of the KPA is that it also can create cabinets from existing IRs as well as Axe-FX cabinets, while the opposite is not possible.  So the KPA can take advantage of new IRs while the Axe-FX cannot take advantage of the KPA’s profiles.

The Kemper profile normally contains the cab element, so you get a great sounding amp (possibly with any number of overdrives in front) as well as the cab. No need for messing with IRs like on the Axe-FX — whereby the IR can have a huge impact on the tone.  So for the Axe-FX, not only do you have to tweak all the parameters of the amp models, you have to choose the right IR cab model.  The Kemper already has this as part of the profile.  You can also pick and choose different cabs in the Kemper (and even create cabs from existing IRs), but it is a well-known fact that many cabs/speakers are optimally chosen to match with specific amps to get a particular sound that everybody recognizes and admires (Celestion V30s for metal, Alnico for AC30s, G12-65 for Dumble, etc.).  So though the Axe-FX gives you the flexibility to choose so many combinations, rarely would you ever use many of those combinations.  Again, the KPA is more usable and practical and doesn’t cater to the features that may be used only 1% of the time. [Update: In 2015 as of FW 3.1.1, Kemper has introduced Pure Cabinet, which is a feature which is a parameter which can be adjusted so that cabinet played through a FRFR system will sound smoother and less harsh — like an amp in the room. Fractal followed suit with a dephase parameter which does a very similar thing].

If your signal chain includes a great mic and mic pre, that is also captured in the profile. For example, if the mic or mic pre add some coloration and distortion, it can be modeled by the KPA. Fractal cannot capture this at all — it must be done by 3rd parties via the Impulse Response.

One interesting piece of info for the KPA. In the Basic manual on page 51, there is a sentence stating about profiling: “However, there is one exception: some distortion pedals use a special design that cannot be profiled accurately, for instance the Tube Screamer.”

I can sort of see why the TS may not be modeled properly, but it is speculation only: I assume Kemper has exhaustively analyzed how amps distort (mainly tube amps) and modeled his box as a harmonic distortion generator. The profiling determines the proper parameters in the algorithm used to create the distortion. However, a TS may distort in ways not taken into account by his algorithm, and thus is not captured during the profiling process. But the EQ and gain may be taken into account as they would be part of the algorithm.

Hopefully in future firmware, he can modify his algorithm to include distortion characteristics from TS-like or other non-conforming devices and include this in a firmware update.

2015 Update: many people are profiling pedal overdrives — many based on the same concept as the Tubescreamer (op-amp with clipping diodes in feedback). I have profiled my own JTK Harmonics Alchemy pedal which utilizes a similar scheme, and the profiling is accurate.

Sound

From basic comparisons on the net, this seems to be a dead heat. You can get great sounds from both.  But the consensus seems to be if you want to get a specific tone, the KPA is better. That is why Fractal introduced tone matching — which is just a fancy EQ method to match the patch tone to some tone from a record. Many VST plugins allow you to do this already.  If you want to get multi-layered, complex tones with superb effects, the Axe-FX may be a better choice.

Conclusion

As I have gotten more experienced, I’ve gravitated to specific sounds and setups. Simplicity is best for me. The KPA fits this. Philosophically, I am drawn towards the KPA because of its modeling technique. It models at a higher level. To be able to do this, you have to understand how and why an amp sounds good, and analyze its distortion characteristics, then determine which ones are measurable and variable.  Fractal doesn’t have to do this at all — they model components in a schematic and use tweaked parameters to make it sound better.  This is a much more involved simulation, and thus why they need to use higher-powered Analog Devices SHARC DSPs for the simulations. Kemper don’t need this as they have analyzed the distortion characteristics from the real profiled sounds and generate the sounds at a higher modeling level  — which is why he only needs to use a Freescale DSP running at a lower clock speed (400 MHz).  It is a much more elegant solution which is not only more efficient, but allows modeling of almost any distortion device.

That is why I think the KPA is the best — for me. Also, the possibility of constantly getting new profiles of amps and sound devices make this device usable for many years to come. New firmware updates don’t mess up existing patches or sounds like it does on the Axe-FX. And the workflow looks like an amp and is so much easier to tweak and use from the front panel. The latter cannot be underestimated. It is so easy and a joy to use to be able to access all the sections with buttons and knobs (especially with the LED indicators on the knobs!). Switching presets, adjusting gain, EQ, and turning on/off effects are all within simple reach.  This is much harder to do with Fractal’s limited interface unless you use their Axe-Edit SW — in which case you need to have a laptop available.

If you love tweaking and need super flexible routing and the best effects, the Axe-FX may be a better choice. Like in a studio.  It also functions as a USB audio device which can save you money. Though most people already have an audio interface for their DAW.  This is why Kemper has not opted to do this — most people already have an interface if they already own a DAW.

If you just like playing specific amps, and getting great tones with no fuss and simplicity — and like to hear new tones without it messing up your old ones — the KPA is for you.

Appendix

Customer Service

Both offer great customer service. However, I have frequented both Kemper and Fractal forums, and my experience over many years is that the attitudes of the founders differ. I prefer Kemper’s friendly attitude towards his customers as opposed to Fractal’s sometimes condescending attitude. 

Marketing

Kemper introduced his unit in 2011. That unit still is the one used today with the latest firmware. Fractal has evolved his hardware and unfortunately no longer supports the older units. This is disappointing as they no longer support the Standard and Ultra models (new amps and IRs are not possible with these discontinued products). Users are hence encouraged to constantly upgrade their hardware ( a la Apple) at significant cost. Fractal could have produced firmware that uses the latest models, but limit effects, or number or amps, etc. and have that available to the older models. But he chooses not to. Because he wants you to upgrade to more expensive models. 

Future upgrades

The great thing about the KPA is that Kemper can continue to evolve the algorithm to make it better, and any changes should be easily handled in firmware.  Of course, if he needs additional signals to create any new parameters from profiling, this could be an issue.  But that would only affect people who want to do profiling. Older units should be able to have the firmware modified to play both kinds. Also, old profiles should be easily backwards compatible as he can create an algorithm which automatically converts to new parameters or simply uses the old algorithm using the old parameters to create the sound.

This is simply much more flexible and takes modeling to a higher-level than Fractal. The latter needs to model each amp component-by-component using a known schematic. And most knowledgeable people know that component tolerances vary greatly and that is why amps don’t always sound the same. I believe Fractal models this variance by introducing many parameters with wide ranges.  This can empirically cover what he is trying to model.  Luckily for Fractal, most people gravitate to only a few kinds of amp models and variations, so he’s able to model them without taking up too much memory space. But Kemper only needs to save the profiled parameter values (not the schematics) — about a few kB —  and use his generator to recreate the sound of thousands of different amp rigs  — some with a very unique sound. This is a much more efficient way (hence the need for a less powerful DSP processor than Fractal). In addition, if you like a specific sound setting of an amp setup that includes EQ/treble boosts and overdrives in front of the amp — the Kemper can model these in the profile!  Again — far more flexible than Fractal’s approach.

Finally — since the KPA is a blackbox distortion generator — it is possible to create distortions that are not possible with any real-world circuit schematic.  Thus, it is possible for Kemper to introduce new distortions and sounds (much like his synthesizers) and create new sounds. Fractal can only model what already exists as a schematic, and thus is much more limiting.

Impulse Responses

Another issue and “snake oil” that Fractal brings up is the use of Impulse Responses (IRs) that exceed 20ms.  The Standard and Ultra use 1024pt IRs (22ms). The Axe-FX II uses 2048pt IRs that now go to 43ms (@48kHz sampling) and also has an ultra-res version that goes up to ~170ms.  Many people have disputed whether you gain anything from this, and I agree. The cab IR is sonically apparent in the first 10-20ms. Anything else after that is most likely early reflections and is an impulse response not of the cab, but of the room.  Beyond a 1024pt IR at 48 kHz (22ms as in the Standard and Ultra models) is really simulating  early reflections and room ambience. This is not what the cab is contributing to. I would rather have the reverb handle the room ambience part.

For Fractal to upsell his “better” Ultra-Res IRs, he needs to be able to process them. Essentially they are just longer-tail IRs. Convolution is an intensive process, but usually is not a problem  for a plugin in a modern day DAW where the IR may be much longer than 22ms.  However, for the Axe-FX, Fractal needs to be able to process these longer IRs without taking so much compute or memory.

Jay Mitchell (a speaker expert on the forums) has mentioned that for Fractal’s ultra-res processing, beyond 2.5 kHz, there is serious loss of information due to decimation in the ultra-res versions:

“To produce an Ultrares IR, the tail beyond approximately 20ms is decimated: lowpass filtered, then downsampled. Based on the frequency content of this tail – it cuts off very sharply at 2500 Hz – the decimation ratio is 8:1. IOW, the sampling rate used by the algorithm that processes the tail is 6kHz rather than the system rate of 48kHz. The demand on CPU goes down approximately as the square of the decimation ratio. This process has been in use for decades, for exactly this purpose (reducing cpu demand). Here’s a page with an overview of decimation: http://www.dspguru.com/dsp/faqs/multirate/decimation.”

Fractal is doing this because he has to with the limited hardware in the Axe-FX. However, in my opinion, it’s really a marketing ploy. This is really useless since this additional IR data beyond 22ms  is not really required for cab sims. Fractal spend a lot of effort trying to market this aspect which isn’t part of the cab anyways.  He’s selling snake oil for his own proprietary IR format, knowing full well that from a scientific point of view, it doesn’t have anything to do with the cab — but the room itself.  But it allows him to sell custom, proprietary IRs which cuts out all the 3rd party vendors that provide great IRs so he can capture more revenue. These Ultra-res IRs aren’t even really high-res — as they are not the true wave file representations of the longer IRs.

For anyone who says this is not adequate — many of the best reverb units use algorithms and sophisticated models rather than strict IR convolution. And they sound as good or better in many applications. This includes Yamaha, Lexicon, Roland, Korg, etc. that have been used on countless records.

After reading more info from the wikPA, Kemper’s comments suggest he uses a 1024pt IR.  The fact that CabMaker can create cab profiles from IRs also seem to verify this.  This is speculation, however, since it’s never been revealed how the cabs are done.  From a size point of view,  for a 1024pt IR, using 24b  (3Bytes), this would take up ~3KB. This is confirmed by looking at wave files of IRs that are 1024pt, 24b. KPA rig profiles files are 3.9kB, so that leaves about 900B of rig parameters. Assuming each parameter has a 1 byte value (range from 0-255), and 50% overhead for recording state information (stomps, cab, effects, etc.), that is about 450 parameters — probably more than enough to cover the amp profile algorithm settings [this is also speculation]. 

There’s been a lot of discussions back and forth between Kemper and Fractal founders.

http://www.kemper-amps.com/forum/index.php/Thread/8982-KPA-IRs-and-math-mumbojumbo/?pageNo=1

Here is one response from Kemper:

“So Cliff is continuing to spread assumtions about our product. I never commented that, I thought it would be ridiculous. Now that he is repeating similar statements, and it makes its way to other forums, It’s time for a statement. I think I have never stated that we do not use IRs. It requires IRs and several other methods to achieve the sound of the Profiler.

If the accuracy was not top notch, you would clearly notice it in the A/B comparison. We were the first to implement such A/B comparison to show how close we get to the original. We have nothing to hide soundwise.

We have two scrambled data blocks in our rig files. Those contain frequency responce information as well as parameters for the dynamics that have been profiled.

Cliff has simply counted the number of bytes without knowing what they respresent, and concluded we use a 256 point IR. As a world class engineer that he is, he must surely assume that there is more information hidden.

I am absolutely with Cliff that a minimum of 1000 points ae required to accurately reproduce a guitar Cabinet. Framing a shorter IR with simple equalizers – as he assumes – will fall short, when strange cabinets are profiled.

A 256 point only IR with equalizers might save some calculation power, would definetely not sound better and would be much harder to handle by code.

I will not reveal details of our methods, but here is a little math if we was using plain IR convolution only:

Our processor is a dual core at 200 MHz. That equals to 400 million operations per second. We run at 44.1 kHz sample rate (44100 samples per second) Divide both numbers:

Result: We have 9070 operations available to calculate one output sample. For calculating a 2048 point IR our processor requires 2048 operations (it’s 1:1). 2048 operations of 9070 operations equals to 22.6% calculation power dedicated to the cabinet IR.

I am with Cliffs statement, that the amp sounds quality is happening in the cabinet by more than 50%. Why would we compromise the cabinet sound, if takes less than a quarter of our processor to make it perfect? A 1000 points IR would save 11%. Thats not a gigant saving. A 256 points IR would require less than 3%. Why should we be that stingy?

Cliff has all these facts in his hand, this was an easy math. He clearly claims that we use less than 3% of the available calculation power for the cabinet simulation, and try to correct the shortcomings with equalizers. Sweet! We dedicate much more calculation power and IR points to the cabinets, how can this be doubted?

I will also respond to another statement of Cliffs, he claimed that the Profiler Amp section runs at 22 kHz only. Of course this is not true either. There is a bandlimiting prior to the amp, by minor reasons. It is not noticable and I might take it out in one of the next firmwares.

Again Cliff could have made an easy assumtion: How could we downsample our amp, while it’s widely known that you have to upsample the amp to reduce aliasing? Every frequency plot on the web made by the Profiler shows that it covers the full frequency range.

It was not my intention to make Cliff look like an amateur, but his frequent bashings over more than a year forced me to publish a clarification.”

CK

—

There’s clearly a rivalry and some back and forth on things. Kemper has taken the high road in all of this while others have not.

Kemper vs Fractal — A Mathematics Analogy

The way in which these two products create guitar tones can be summarized in a mathematical way. Take for example, the guitar input as g(t) and the amp+cab output as a(t) where t is time.  There is some transformation function K such that a(t) =  K{g(t)}.  The amplifier in some way is transforming the guitar signal into the signal we hear from the amp+cabinet.  K clearly involves EQ, spectral density, distortion, and other transformations in both the time and frequency space.

The issue is then — how do we arrive at this function K ?  Well, there is already a known way — and that is from looking at the amp itself — through the circuit schematic.

Fractal Way for Determining K

The circuit schematic is a roadmap for how electrons traverse through the components and react in a specific way. This schematic is one way to determine the K function we are looking for, and is arrived at by modeling how each component reacts under voltage and current.  Each component of an amp has a voltage and current characteristic over time, frequency, amplitude, etc.  This can be independently modeled, then put together in a schematic and simulated so that the guitar input g(t) applied to it will result in the amp out a(t).  If you really wanted to model accurately, you would need sophisticated SPICE modeling (based on device physics) of each component and take into account all parasitic capacitances, inductances, and secondary effects.  This is what is done in very advanced ASIC analog and digital circuit designs for semiconductor chips. However, this takes a lot of compute power to be able to do this in real time with low latency. At this time in 2014, a fully-loaded Intel i7 is not able to do this.

So how is this done without such massive compute power?  This issue has been in simulation of electronic circuits for a long time, and the solution is to model each component at a behavioural level — i.e. understand its behaviour by looking at higher-level behaviour rather than at the device physics level.  You extract specific characteristics of the device under varying conditions, and model that using simpler equations rather than the device physics equations. This is less complicated and thus the compute time drops dramatically.  You can still choose to use physics modeling in areas which don’t involve a lot of compute, and use known methods for solving circuit equations to arrive at the answer for a(t).  The key is to try and do this as efficiently as possible so it can be done in real time with less compute power (and hence a cheaper chip).

Since Fractal needs an amp schematic to model an amp, I believe this is the method he uses to determine K.  Since the compute can be intensive with many components and parasitics involved, he needs a fast processor like the ADI TigerSHARC. In fact, as his modeling tries to model the components and secondary effects more accurately, he needs more power, hence the need for dual-DSPs like in the Axe-FX II  — one for effects and one for amp modeling (previous Axe-FX Standard and ULTRA models use only one DSP processor).

As he continues to tweak his models, he may introduce or discard empirical parameters to get better component models.  Whether this is determined by comparing to actual voltage/current data or by simply listening is not known.  But in general, this method relies almost entirely on the amp schematic and the accuracy of the modeling of the components within that schematic, and using circuit simulation to arrive at the function K.

As we all know, not all amps sound the same — even though they are supposed to have identical schematics and components. This is because each amp has unique values due to component tolerances and parasitics. Fractal cannot model these except to be able to introduce parameters which vary the component values. This is why I believe he has many tweakable parameters — to take into account component tolerances, parasitics, etc. These small changes can and do lead to changes in tone which for some amplifiers, are needed to achieve their signature sound. This is why it may be necessary to tweak on the Axe-FX platform to get the tone you want.

Kemper Way of Determining K

Kemper takes a very different approach, which isn’t hard to understand once you know that he also designs world-class synthesizers. His approach is to start from scratch. To find K, his assumption is that you only need to know a(t) with a corresponding g(t) input. He doesn’t use the electrical schematic used to generate the real-world tone.  Here is the key: Kemper has previously analyzed what a majority of different amplifiers do in real life  (K function), and has determined an algorithm which mimics this behaviour.  These might be EQ, spectral, distortion generation of different kinds, phase shifts and amplitude, etc.  So he models the amp’s K function like a blackbox — a generic harmonic distortion generator if you will.  This is not surprising, given that he is also the inventor of the groundbreaking Virus line of synthesizers.

In the real world, before the advent of digital signal processing, there was a need to amplify the guitar — using analog circuits with these components:  tubes, resistors, capacitors, and transformers — all connected together to create gain and resultant distortion.  However, in the DSP world, this method of arriving at K can be circumvented by directly looking at the output vs input.

Kemper’s approach takes the behavioural modeling approach to a higher level by modeling the entire amp this way. Fractal may only do this for the components, but still uses the schematic and electronic circuit simulation to arrive at the K function. Kemper dispenses with this and attempts to model the whole box by observing the output behaviour given a specific input.

The key in this strategy is determining what transformations are done on the input signal to produce the output. Once this algorithm is determined, you need to be able to measure the parameters within the algorithm which makes each amp’s K function unique.  This is the profiling part of the Kemper. And because the algorithm takes input parameters to do the modeling (and they are the same generic parameters for each amp profiled), the profiled data is very small as it only represents the input parameters to the distortion generator.  

Because the K function is determined by profiling a specific amp, if that amp has a sweet spot with a special sound, it can be captured.  Fractal may not be able to capture this because that model is based on a schematic and component values/models which may not take into account any secondary interaction or values/settings which the amp exhibits. Kemper’s method simply does not care — it just profiles the amp to determine the K function and you have it. Not only that, the entire signal chain is profiled. That includes the cabinet, mic, and mic pre. Many people also add boosts or overdrives at the input of an amp, and that is also included.  Kemper has also devised a way to separate out the cab effects from the signal chain, so that each profile allows you to use the cab effect in another profile. The profiling step thus captures the amp and cab IR in one step.

Modeling Efficiency

Kemper’s approach is more efficient as it requires less compute to arrive at a K function which may be sonically identical to the amp.  Fractal arrives at a K function, but takes a much more complicated and compute intensive route to get there via circuit simulation.

If we assume that every amp could be modeled by one specific circuit, then the only thing that would change would be the component values — and that would make Fractal’s modeling much easier. Taking it one step further in fact, would bring it conceptually closer to what Kemper is already doing.

The genius in Kemper’s approach is that he is able to determine an algorithm which can model thousands of different amplifiers, and determine their K function by profiling them using a set of signals which determine the input parameters to his algorithm — and hence yield the K function.  Fractal cannot do this as the K function for each amp is determined by the schematic, which has to be modeled in the firmware. Hence Fractal are limited to how many schematic models are in his firmware.  Also, changes in firmware to make modifications to component parameters, etc. could easily upset existing patches for amp sounds. In the Kemper, he can continue to make changes to the algorithm if he chooses to, and still keep the existing parameters, thus making all previous amp profiles backwards compatible.  Since his method of determining K is at a higher abstracted level, changes in profile parameters could be more easily handled. Fractal on the other hand, has a more difficult time with this, as most of his modifications are done at a component level, and thus are not easily backwards compatible with older modeling parameters (he will sometimes add parameters or remove them completely). This has been very apparent by looking at users’ responses to each new firmware upgrade. Each new firmware release has “better” modeling of components which could drastically change the sound of existing patches — making tweaking or recreation of the patches necessary.

Conclusion

I like Kemper’s approach because it is more elegant and more efficient. The method can be easily modified to add additional parameters to profile, yet still maintain previous parameters for the algorithm — so potentially no change in sound and backwards compatibility. Since each profile is represented by values of parameters input to the algorithm to generate the K function, the profile itself can be quite small (4KB). Finally, since the only thing required is an actual amp or distortion device to generate the K function, the Kemper Profiling Amplifier can determine the K function on a near limitless number of devices — amps, some pedals, etc. and at specific “sweet spot” settings.  Fractal can only model based on a schematic and component behaviour which is fixed within his model framework.  Though there are a lot of parameters to tweak and a large number of models, it still cannot come close to the huge number of profiles available in the Kemper. Finally, the Kemper profiles can be shared and distributed — making available huge numbers of amp tones available.

Side Note

One thing not mentioned often regarding the Kemper —  are the actual definitions of the parameters within the profiles. Since this is a binary file and proprietary, it is not known.  Since Kemper has access to these profiles and parameters, you can imagine what information can be gleaned from this. For example, they can look at a set of profiles of the best VOX amps, or best Marshall, Mesa, Bogner, etc. and analyse the profile settings to start to understand why some amps sound better than others, or what characteristics define their sonic signature. Then they can go in and tweak the profiles themselves to optimize their tones.  Only Kemper can do this since only they know the profile parameters definitions and algorithm.  But if they ever chose to do this, I can only assume that the optimized profiles coming from the extensive datamining of existing profiles may result in fantastic sounds — and perhaps some that have never been achieved before.

As a future product, I would love to see software be able to manipulate the profile parameters — in real time — to allow users to optimize or completely change any profile.  I am sure some extremely innovative and new sounds could be created this way.

Workflow Revisited

The KPA has another big advantage over the Axe-FX : usability.  The layout of the front panel exactly matches most guitarists’ workflow (including mine). It is so much more intuitive because you have access to a larger display, LED indicators and knobs — like on real amps, pedals, etc. Though there isn’t a software program that mimics the display (like Axe-Edit on the Axe-FX), you don’t need it because it can all be done from the front panel (like on a real amp or pedal).  To tweak the Axe-FX, you need to go into endless pages of menus and parameters. And with only a few knobs, you really need a computer running Axe-Edit to do this properly. And Axe-Edit functionality is no longer supported for the Axe-FX Standard or Ultra — only on the II and later.

For a MIDI controller, I previously used a Behringer FCB1010 with the Axe-FX. The FCB1010 is the most most economical MIDI controller pedalboard out there. Though Behringer has a reputation for cheap, low-quality products, this product has been road-tested by many guitarists and is even used by professionals on tour. It is very rugged and reliable.

With the Axe-FX, the FCB1010 was cumbersome to program, but it did work using the Uno EPROM and some other software to program it.   Fractal provided no support for this, as they offer their own, extremely expensive MIDI pedalboard (MFC-101) which is $750 (recently dropped to $650 in October 2015, most likely due to the AX8 introduction). In comparison, the FCB1010 is about $150 ! I think it is ridiculous to spend that much for a MIDI controller unless you are a touring pro.  You can buy a good modeler (used Axe-Fx Standard or Avid Eleven Rack) for that much!

Using the UNO4Kemper EPROM and labels, the FCB10101 is even easier to use with the Kemper, and the features — such as automatic tuner when the volume pedal is up, or the display that works with the tuner — are brilliant!  The Kemper engineers directly worked with the UNO4Kemper team to offer support for these features. Fractal did not choose to work with the UNO4Kemper team for FCB1010 support. Not surprising as they sell a competing over-priced MIDI pedalboard.

Overall, my experience with the FCB1010 (UNO4Kemper EPROM) + KPA versus FCB1010 (Uno EPROM) + Axe-FX Ultra — there’s no comparison. The KPA combination wins hands down for usability.

And with the KPA rig manager with a laptop connected to the KPA using USB, I can preview and browse through thousands of profiles quickly and easily.  There’s no need for a software program to actually edit the KPA because the front panel is so easy to use.

Kemper has introduced the Kemper remote footcontroller, and this enables a few features on the KPA — most notably the looper. This is a welcome addition to the family, though at 400+ Euro, it is still a high-priced item for what it does. Both Kemper and Fractal seem to overcharge for their foot controllers in my opinion.

Profiling

I finally setup a signal chain to profile a stomp OD pedal. Kemper says that the tubescreamer – like pedal is hard to profile, but I was able to do it with a JTK Harmonics Alchemy pedal into a BadgerPlex trilogy (EQ shaping) into a Pearce G2x amp (channel 2 clean) into a THD 2×12 with Celestion V30s, then mic’ed with an SM57 into a tube mic pre then into an audio interface. Within the DAW, an EQ plugin was used to enhance the tone.

Amazingly, the profiling was very simple and the resulting sound was very close to the original. You can really tailor the sound with the amp parameters as well as the EQ, gain, etc.

I was pleasantly surprised and was amazed that the profiled tone was close. It had the same character and timbre.

This alone makes the KPA capabilities far beyond the Axe-FX.  I can now profile multiple amps, overdrives, and software sims with any number of amazing VST plugins that emulate great mic pres and EQs.

Update: 8/17/2015

On 8/10/2015, KPA introduced “Pure cab” feature which is an adaptive filter to remove phasiness and smooth over artifacts in the IR.  It is an effect that can give a good sound to a cab — maybe close to the “amp in the room” sound without actually doing that. A day later, Fractal announced a similar feature —  “dephase” parameter.

The battle continues …

Once the Line6 Helix becomes readily available, this will make things even more interesting as the GUI and usability on that unit seem superior to either Kemper or Fractal. And it’s in a footcontroller chassis complete with an expression pedal.

Update: 10/4/2015

Update on Fractal AX8. This is a floor version of the Axe-FX II but appears to have only one stereo path and one amp/cab.  Price is very competitive at $1399 USD. Units are supposed to start shipping at the end of October 2015. It will have all the amps that the Axe-FX II has, but only one amp/cab instead of 2.

Update: January 2016

AX8 is still not available to many who have been on the waiting list since October 2015!  I speculate it was pre-announced early because Line6 Helix forced their hand, and Fractal didn’t want people jumping ship.  Regardless, it’s poor communication to announce in October and still not be shipping.

Update:  1/24/2017

The AX8 has dropped in price to $1299.99 since late 2016, but that’s because it is now being made (i.e. assembled) in China (most electronic components are made in China or Taiwan already).   The Kemper firmware 5.1 has stunning new delays — dual and quad structures with pitch, crystal, and frequency shifting in the loops.   And Line6 Helix is offering a VST version soon.

These are fun times, and customers surely will benefit from all the competition!

Kemper Profiling Amplifier: http://www.kemper-amps.com/

Fractal Audio Axe-FX: http://www.fractalaudio.com/

Line6 Helix: http://line6.com/helix/