Impulse Responses (IRs)
From Axe-Fx II Wiki
The information on this page supplements the official manuals.
About Impulse Responses (IRs)
An Impulse Response (IR) is a collection of data representing sound measurements taken from a speaker cabinet or system. A test signal is played through the actual speaker, recorded, and used to generate a profile.
The Axe-Fx II and AX8 use IRs in the Cab block to reproduce the measured response and emulate a particular speaker cabinet, as well as for modeling of microphone types.
The terms "cab", "user cab" and "IR" are often mixed up.
"An IR stands for "Impulse Response". In mathematical terms it is the time response of a system to a Dirac delta function (also known as an impulse). An IR can be used directly as the coefficients for an FIR (Finite Impulse Response) filter. In the modeling world IRs are obtained from real speakers and when processed using an FIR filter produce extremely accurate results. In essence an IR is a "sample" of the speaker and microphone and uses very similar principles. However the quality of any IR is subject to the talents of the individual(s) capturing the IR. Mic placement, preamp choice, etc., etc. are important as you are essentially recording the speaker. In the old days modelers used EQ to emulate speaker response but I don't think there are many left that still use that technique. So the quality of the IR is really the issue here. The original Axe-Fx pioneered this technology which has since become almost ubiquitous." source
Impulse Responses: supported by which Fractal Audio products?
- Axe-Fx II: yes
- AX8: yes
- FX8: no
Cabinet block and IRs
This wiki page is about Impulse Responses in general. For information about the Cabinet block and its parameters, visit the CAB block page. The Cab block is the container that processes IRs (built-in and external ones) and provides additional parameters.
Near-field and far-field IRs
Most speaker IRs represent the tone of a speaker that was recorded with the microphone close to the speaker, aka "near-field" or "close-miked".
"Far-field" IRs represent the sound of a speaker that was captured at a longer distance. There are a couple of far-field IRs among the stock cabs, created by Jay Mitchell ("JM"). These IRs are better suited for recreating the "amp-in-the-room" sound of a traditional guitar speaker.
IRs and FRFR amplification
Read this: FRFR and close miking (versus amp-in-the-room).
Length or resolution of IRs (Ultra-Res, Hi Res, Normal Res)
Fractal Audio devices and software support IRs of various lengths, measured in number of samples (points) and milliseconds:
- Normal: 1024 samples, 20 ms. You can often use Normal Res without a noticeable impact on the tone, compared to Hi Res. The Cabinet block's Stereo mode supports two Normal Res IRs.
- Hi/Ultra: Hi Res IRs (2040 samples, 40 ms) and Ultra-Res IRs (up to 8000 samples, 170 ms). Hi Res and Ultra-Res IR processing requires more CPU power than mono or stereo Normal Res. But Ultra-Res is more efficient than Hi Res, which results in about 4% less CPU usage AND higher resolution! The Cabinet block's Stereo Ultra-Res mode supports two Ultra-Res IRs. Non-Ultra-Res IRs will be processed in stereo mode as Normal Res. If one IR is Ultra-Res and the other not, then the Ultra-Res IR will still processed as UR and the other as Normal in stereo mode.
Ultra-Res speaker IR processing is a proprietary technique that enhances the spectral resolution of an IR without adding CPU burden or storage requirements.
Ultra-Res IRs do not support size warping, which is why the Cab block's Speaker Size parameter is unavailable in Ultra-Res mode.
Ultra-Res lRs are displayed in italics in the editors.
Fractal Audio (source):
- "The problem with conventional IRs is that they are too short to capture the detail in the low frequencies. There are those that maintain 20 ms is the maximum length you need to fully replicate the speaker. This would be about 1000 samples at 48 kHz. I disagree with this as I have many IRs here that exhibit significant energy beyond 20 ms. I believe the room has some influence as the low-frequency modes of the room will impact the resulting sound. The amount of this impact depends on the room, the mics, distance, etc., etc. Or perhaps certain speakers have particularly high Qs in the low frequencies. Regardless, it is my opinion that you need IRs much longer than 20 ms to fully capture the "mic'd amp in the studio" sound. My tests show that IRs of 8000 samples are required to fully capture the low-frequency detail. Unfortunately to process an 8K IR in real-time require copious processing power... Fortunately I have developed "Ultra-Res" cabinet modeling. Ultra-Res cabinet modeling provides the frequency detail of a very long IR with little or no added processing power requirements. The following image depicts the response of Ultra-Res cabinet IR processing: ..." (see thread)
- "Existing IRs will still be processed as usual. Ultra-Res IRs will be tagged as such which will indicate to the processor to use the new processing algorithms. Note that Ultra-Res IR data is not conventional IR data."
- "The frequency resolution of an IR is the sample rate divided by the number of samples in the IR. The window function has nothing to do with frequency resolution (except for making it even less). So a 1K IR at 48 kHz sample rate has a frequency resolution of roughly 48 Hz. If a speaker has a resonance (formant) at, say 80 Hz with a Q of, say, 3.0, then 48 Hz is insufficient to capture that resonance accurately. You need a frequency resolution of several Hz to accurately recreate that resonance. I chose 80 Hz and a Q of 3 because that's what that response looks like. The Q could even be higher than that. It doesn't take much mental energy to realize that if you have a narrow formant at a low frequency then you need fine frequency resolution to reproduce that. An 80 Hz formant with a Q of 3 only spans about 25 Hz. Obviously a frequency resolution of 48 Hz is not going to be able to reproduce that. Windowing only smooths the response even more. This is basic FFT theory. The less time-domain information you have, the less frequency domain information you have and vice-versa. This is the uncertainty principle. I always window IRs with a Hann window."
- "Another way to look at it is to think in terms of formants. That particular speaker has a pronounced 80 Hz formant. It takes well over 100 ms for the energy of that formant to decay to the point of imperceptibility. Obviously a 20 ms IR can't reproduce an event that occurs for over 100 ms. Here is a zoom of the original non-minimum-phase IR (IOW raw time response)... (see thread). You can clearly see the 80 Hz formant. There are some room reflections but they are very small. The 80 Hz formant starts well before any reflections. It's obviously a high-Q resonance as it rings for quite a while. The higher the Q, the longer it takes to decay."
- "Here's another example. (see thread) This is one of the new OwnHammer IRs. The IR is OwnHammer_412_MAR-CB_D-120_SS_RBN-121. These IRs are 100 ms long (4800 samples). I windowed the original IR to 4K to prove a point. The blue trace is the original IR (windowed to 4K samples). The green trace is the "typical" 20 ms IR (windowed to 1K samples). The red trace is the Ultra-Res version."
- "The problem is that human perception is logarithmic and IRs are a linear process. 48 Hz resolution is way more than necessary at, say, a few kHz but not nearly enough at low frequencies. The brute force solution is to use very long IRs, 8K or more. Ultra-Res solves this in a novel way that uses little to no extra processing power and no additional latency."
- "Normalization is your friend. Rectangular windows are simply truncation and are generally regarded as bad practice due to extremely high sidelobe levels. The choice of window is subjective. I actually use my own custom window that is not really a Hann window but that's proprietary information. My window preserves more frequency detail while still suppressing Gibbs phenomenon. Windowing trades off frequency resolution for sidelobe suppression. My window is optimized for the unique statistics of IRs. For a random process I tend towards Bessel-Kaiser windows. IRs have unique statistics that aren't addressed by any of the standard textbook windows."
- "It is desired that the IR be 8K samples or more."
- "Let me state these points:
- We don't record guitar amps in airplane hangers or anechoic chambers. We record them in studios.
- When we record a guitar amp we carefully set the amp up in the studio to get the best sound "on tape". This involves moving the amp around, placing gobos, etc. When we collected the Producer's Packs IRs we spent hours arranging the amps/speakers, mics and gobos and playing through the amp and readjusting until we were satisfied. This also included adjusting the preamps and mixing board. In one studio we found that we got the best tone raising the cabs off the floor by a couple feet, orienting them towards a particular wall and placing gobos behind (this was the engineer's standard recording arrangement).
- At this point our objective of the IR is to capture the sound of that amp/speaker at that position in the room, with the gobos, mics, preamps, etc., etc. The goal is not to capture the raw sound of the amp/speaker in an airplane hanger or outside using a ground-plane measurement and measurement mics. That might be someone else's goal but it is not ours. IOW our goal is to treat the cab, mics, preamps, room, etc. as a whole, as a good engineer/producer would.
- Subsequent analysis of the data shows that there is significant energy out to 100ms and even beyond. However there is little energy beyond 200 ms or so (as it should be in a well-designed studio). This observation was the catalyst for the Ultra-Res algorithm. There are other observations about the statistics of the data that I cannot disclose.
- Some cabinets displayed noticeable resonances at low frequencies. Others did not. The frequency of these resonances were not consistent and, not coincidentally, matched the measured resonance of the impedance sweep. It is a logical conclusion, therefore, that the resonance was NOT caused by the room but by the speaker/cabinet combination. Furthermore a plot of the group delay for the raw data showed that the delay of the resonance was too short to be a room mode. Regardless, whether the resonance is from the speaker or room or mics or preamps is irrelevant. All we care about is recreating the sound of that speaker as it would be recorded as accurately as possible.
- Truncating an IR destroys information by definition. We don't care where the information comes from, be it the speaker or the room or the mics or the preamps. We want all the information. If a plot of the frequency response of a truncated IR differs considerably from the non-truncated version then we have lost information and concomitant accuracy.
- NO ONE producing commercial IRs records them in an airplane hanger, for obvious reasons. The best ones are done in a studio using the same technique we used for the Producer's Packs: setting up the cab, adjusting the position, mics, preamps, etc. and playing through the amp/cab and readjusting until the best tone is achieved. The new OwnHammer IRs are an example of this. Many, if not all, of those IRs exhibit significant energy to 100 ms (and likely beyond but the data stops at 100 ms). Truncating them to 20 ms destroys vital information. You can argue the semantics all day long. I've compared truncated and non-truncated and the difference is clearly audible. It is especially noticeable when chugging power chords. You can hear the resonance. It goes "bonggggggg" as opposed to "thuk". Most importantly it sounds "better" IMO.
- Ultra-Res is an algorithm that markedly increases accuracy. It gives the frequency resolution of a 200ms IR without additional processing overhead and no added latency."
- "Ultra-Res is especially powerful in Tone Matching applications, particularly real-time matches and was another impetus behind the development."
- "The myopic only see the IR as a capture of the speaker's "unadulterated" response. As I stated before I believe the future is treating IRs as capturing the entire recording chain including mics, preamps, etc. and have pushing in that direction. We have already seen the fruits of this labor in the Producer Pack and OwnHammer V2 IRs. We used mainly PP and OH IRs at Axe-Fest this weekend and the results were stellar. Andy Wood's tone was among the best guitar tones I've ever heard live and we dialed it up in 10 minutes under far less than ideal conditions. It consisted of the Two Rock amp model and the EV 12L Mix IR. When you include more than the speaker response in the IR you can have low-frequency resonances that persist for tens of milliseconds or more. Truncating an IR destroys this LF information. In many cases this LF information loss would probably not be perceptible. In other cases, from experience, it can be extremely noticeable. The bottom line is that you can always remove the information if you don't want it but you can't add back what isn't there."
- "Let me phrase this another way. An IR can consist of the "raw" speaker response plus none, one, some or all of the following: mic, preamp, room, power amp (e.g. you want to capture the response of a tube amp driving the speaker), etc. If you only care about the raw response then a short IR is all that is required. However if you want any of the other elements as part of the IR then a longer IR may be necessary. Ultra-Res gives you the OPTION of processing longer IRs."
- "If the .wav is only 40ms long there is no sense in converting to Ultra-Res as you won't gain anything. Over 80 ms is desirable. The maximum length supported is 170 ms or so. Anything longer than that is truncated to 170 ms." source
- "To get the optimum results the length should be 170 ms or more. As the length gets shorter you'll lose information. However there may not be any information to lose. It all depends on the IR. I've seen long IRs where only the first 100 ms or so is actual information and the rest is silence. OTOH I've seen 100 ms IRs where there is obviously more information but it got truncated. You lose nothing with Ultra-Res except the ability to change the size of the cabinet. You gain better sound and less CPU." "You can't mix Ultra-Res IRs as the data is not compatible. However... we foresaw that and the UltraRes conversion process produces two files: a .ir file and a .syx file. The .ir file is the raw IR data that can be imported into CabLab for mixing purposes. So CabLab can take .wav, non-Ultra-Res .syx and .ir files as input to the mixer section and product Ultra-Res .syx files." "The .ir files are included with our cabinet packs. We will not be offering .wav files. If you have the .wav file you don't need the .ir file. A .ir file can ONLY be used with CabLab. If you use the Axe-Fx II to capture IRs it will only generate .ir and/or .syx files. No .wav files are generated. The resulting data can only be used on Fractal Audio products." source
- "It depends on the IR. Ultra-Res improves low-frequency resolution. It is very apparent with some IRs and virtually inaudible with others. It all depends on the low-frequency formants in the original IR. If there are significant, high-Q formants Ultra-Res will preserve those whereas conventional, short IRs will not. Audibility also varies with the amp being used. The difference is more audible with high gain as this will excite the formants more. Low-frequency formants vary with the type of cabinet and speaker. Some cabinets have a smooth low frequency response. Others have prominent formants. The mic also has an impact. Some mics will accentuate the formants. The room also contributes if it has strong LF modes. Furthermore some people like to capture an IR using a tube power amp. In this case you WILL get a significant formant at the low-frequency resonance of the speaker. A conventional IR will not capture that as the Q of the formant will exceed the resolution of the IR. Ultra-Res will capture that formant as Ultra-Res has 8 times the low-frequency resolution. Those who claim they can't hear a difference are correct. They can't. It's nothing to be ashamed of. But because they can't doesn't mean others also cannot. I can clearly hear the difference but I've trained myself on what to listen for. I vastly prefer Ultra-Res and only use Ultra-Res IRs in my personal patches (aside from the TV Mix, which is just a magical IR)." source
- "The length of the sweep only determines the signal-to-noise ratio. If the room is completely silent the sweep can be infinitely short (an impulse). To overcome ambient noise you need more energy in the applied stimulus. With an impulse you can only increase the power so much before the amplifier or the speaker or the mic or the preamp, etc. distort. However, if you spread that power out over a longer period of time you can increase the energy and therefore increase the SNR. Think of it this way: a 1 ms pulse at 1000W has the same energy as a 1 second pulse at 1W. Now you can't just put a 1 second pulse into a system because the pulse has little frequency content. A 1 second sweep over the band of interest allows the transfer function (IR) of the system to be obtained via deconvolution. There are other signals you can use like pseudo-noise and MLS sequences but a "chirp" has the best characteristics. In the early days of room IR capturing they used impulses generated by popping a ballon, firing a starting pistol or clapping two boards together. The results were poor due to low SNR. This lead to the development of signals that have higher energy. To get the IR of a room long sweeps are typically used because there is a lot of ambient noise and the "returned signal" is weak (the reverb portion of the response is very low compared to the direct signal). When close mic'ing a speaker the ambient noise is low and the signal strength is very high so a short sweep is adequate. In fact you could probably get away with 100 ms or less in a studio environment." source
- "I've never seen a cabinet IR (and I've examined thousands) that has any significant content beyond 150 ms or so. Most cab IRs are under 40 ms. The exception to this would be a "room IR" where the mic is very far from the speaker and the room is significantly reverberant. But one wouldn't normally use that as the primary tone, instead to add a little ambience to the tone and the loss of information would be imperceptible in context. Modeling products typically use IR lengths of 1K samples as this covers 90% of IRs ever captured. We support 2K and Ultra-Res (which is equivalent to 8K) which covers 99% of IRs. The amount of CPU power required to process an IR is proportional to the length of the IR. To support a 500 ms IR (24,000 samples) would require over ten times the CPU power of a 2K sample IR. It also requires over ten times the memory for storage. Given that that vast majority of IRs do not have any information beyond 40 ms it is wasteful of CPU and memory resources to support IRs longer than 2K." source
(about Tone Matching and Ultra-Res)
- "In Realtime mode the raw internal IR length is 8K which you can dump." source
- "You can export the Tone Match to CabLab and create and Ultra-Res IR." source
Fractal Audio (source):
- "I'm a huge advocate of longer IRs. In fact I think I was the first to advocate it despite all the naysayers. I pushed OwnHammer (and others) to increase their IR lengths and they were the only ones who acted on that advice (so far, maybe the other guys will start to follow suit). Ultra-Res was born out of the desire for longer IRs.
- For recording you don't need to use the cab block in the Axe-Fx though. Record the raw amp sound and then "re-cab" it later. This way you can try different cabs. Cab-Lab is great for this. Cab-Lab does not do Ultra-Res processing. It creates Ultra-Res files for the Axe-Fx but it does all processing at the full IR length up to 8K samples. You can use other convolution plug-ins as well.
- The reason for Ultra-Res is that long IRs have several drawbacks:
- They require lots of storage space. Not an issue on a computer but on a hardware product that means expensive non-volatile memory.
- They require lots of processing power if you don't want any latency. On a computer it doesn't matter since latency is a non-factor if you are processing prerecorded tracks. On a hardware product we must have zero latency.
- So Ultra-Res was devised as a way to exploit the statistics of the data to give the benefits of longer IRs without the usual hardware drawbacks.
- In my tests I've found that 8K samples (170 ms) is more than enough. I think 500 ms (24K samples) is overkill and if an IR has significant energy out that far then it has too much room in it. The speaker and cab itself are never more than 100 ms, usually much less. Anything beyond that is the room. I personally don't like IRs with lots of room in them. A little bit of early reflections are nice and make things sound less direct but too much room makes the sound get lost in the mix.
- There's no meaningful data beyond 150 ms and if there is, it's the room and you don't want that much room."
- "No big deal, just some improved processing algorithms. The UltraRes cabs in Quantum 2.0 were all reprocessed with UltraRes". source.
- UltraRes 2.0 is the next level of evolution for our patent-pending speaker simulation technology, with even greater accuracy than the original version. UltraRes 2.0 cab files are backwards compatible with previous Axe-Fx and AX8 firmwares supporting UltraRes 1.0.
"1" from the speaker is the near field. The response of a speaker in the near field is very different than the response in the far field. In the near field the response changes (drastically) across the face of the transducer. Even moving the mic a fraction of an inch will result in a very different sound. 10 ft. from the speaker is the far field and the response changes smoothly as you move across the field. If the near field were the same as the far field then the sound wouldn't change as you moved the microphone and you could place the microphone anywhere on the face of the speaker. Anyone who has mic'd a speaker knows that this isn't the case." source
Minimum Phase and Auto Trim
If desired, the phase of an IR can be manually adjusted or automatically. There are two ways to have the software do it:
- Min Phase: the IRs will be time-aligned.
- Auto Trim: superfluous data at the beginning of the IR will be removed.
Getting the phase right is important when mixing IRs together. When neither Min Phase or Trim has been applied, the IR is "raw", containing the original phase details.
Create your own Impulse Responses with IR Capture
Read this: IR Capture.
IRs compared to Tone Matching
"Tone Matching is a nifty feature and certainly useful but you'll get far more satisfaction by concentrating on capturing good IRs. The single most important aspect of recording guitar amps is micing the amp. Therefore the single most important aspect of using your Axe-Fx is the IR. People are too hung up on "matching" or "profiling" an amp but fail to realize that when you are doing that you are basically capturing an IR. If you capture the IR separately now you have an IR that is fully separated from the amp and therefore can be used with all models. Matching and profiling cannot mathematically separate the amp's frequency response from the cabinet frequency response. Once you do this you'll be surprised at how accurate the amp models are. I do this all the time and find Tone Matching is unnecessary now (in fact many of the amp models have had their built-in matching data removed in the latest firmware). Any differences between the model and the real amp are so minuscule as to be immaterial. A little tweak of the tone stack or EQ is usually enough to remove and differences. Besides, once you get into mixing you'll realize that you'll be applying EQ anyways so tiny differences in EQ are irrelevant. Moving the mic just a small amount drastically changes the sound. The best producers have mastered micing. You can only fix so much via EQ since EQ is essentially painting with a broad brush where mic technique is akin to using a fine-point brush." source
IRs of acoustic tones
To emulate acoustic instruments (acoustic guitar, cello, violin etc.), an IR of an acoustic body can help. You can find some here.
Acoustic sounds benefit from long IRs, so Ultra-Res IRs are preferred.
Differences between IRs for Axe-Fx II Mark I/II/XL/XL+
"The XL has a different sysex ID and therefore requires different cab files. The only difference in the files is the sysex ID.” source
Commercial and free IRs
The Axe-Fx II and AX8 include a lot of stock cabs (factory cabs). You can also create your own IRs or get additional ones from the sources below:
- Fractal Audio Cab Packs
- ML Sound Lab
- Red Wirez
- Dr. Bonkers Soundlab
- forum thread
- free Ultra-Res IRs
Axe-Edit's and AX8-Edit's built-in Cab-Manager lets you load IRs into the hardware.
Fractal-Bot saves and loads entire banks.
IRs for Axe-Fx Standard / Ultra
IRs for the Axe-Fx Standard/Ultra must be converted to be able to use these with the Axe-Fx II. source
It's no use converting 1024-point IRs to 2040 points because they don't contain the necessary data. You need an original WAV-file of sufficient length to create a 2040 point IR.
Information about guitar speakers
- Speakers: the final frontier
- Legendary Tones: G12M versus G12H
- TGP: EVM 12L versus EVM 12S
- Sound On Sound: Choosing guitar-amp speakers
- Sound on Sound: Understanding and recording guitar speakers
- Guitar Player: 15 12" speakers
- Rivera: 15 speakers compared (YouTube)
- Eminence tone guide
- Fenderguru: Selecting speakers
- Guitar Player: Alnico Taste Test
- Celestion history
- Pre-Rola greenbacks explained, part 1
- Pre-Rola greenbacks explained, part 2