DiyMobileAudio.com Car Stereo Forum banner

1 - 20 of 29 Posts

·
Registered
Joined
·
4,939 Posts
Discussion Starter #1
Some time ago I made this little guide how to measure your system. I'd like to continue this with a more complete guide how to finalize the setup to get the most out of your system. This will include about everything under the sun because I believe it's important to understand the large picture. For those who haven't read the guide it's available here:

Measure your system.

Before I start, this thread should be seen as a compliment to the other one. I'm not a big supporter of the "setup your entire system by ear-camp" but I've often said you need to do some parts by ear. Soo, what's up with the fine tuning I talk about? Isn't it enough to tune to a pre-determined target response and leave it at that? What do we hear, what are we aiming for? These are the kind of questions I'd like to discuss. I'm not a know-it-all but over the years I've made a few conclusions what's important and what's not. Hopefully someone can contribute further in this subject to make it an interesting read.

Just as the other thread I'm going to split my posts in several parts as I will discuss several subjects. Lot of the discussion will involve measurements but it will also include subjective stuff this time. I often refrain from using subjective terms when describing audible performance of speakers/equipment or systems because I believe they are completely useless. I'll explain why later on. I'll also try to discuss some of the terms often used in the audiophile industry because I think noone really has touched the subject. Anyway, I'll share some of my thoughts in the following subjects:


* Tonal balance - Target response curves and acquisition.
--- Subgenre: Source material, recordings, compression.

* Staging - The importance of Equalization and Time Coherency.

* Advanced measurement data acquisition for troubleshooting.

* (Install) Speaker placement - Pros and Cons, the best compromise?

* The importance of 'Distortion', how you can optimize performance.

* Speaker choices - What should you look for?

* The human factor - Listening fatigue, audible correlations with what we can measure, are we hearing the same things?, subjective terms - what do they mean? (large subject, saving this for last).


----

----
 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #2
Tonal balance - Target response curves and acquisition.

Ok. The first subject, this one is almost like beating a dead horse by now. Those not experienced with car audio is often under the impression that the response should be flat for best tonal accuracy, after all - a large group of the home audio people seem to believe it's the holy grail of tonal balance. In fact the system I listening to right now while writing this, got a quite flat response if we would smooth the plot (this is a bandpassed measurement);



This is a system that I consider to sound very balanced and accurate with a little high frequency "sparkle" - nothing too overemphasized. Several people who have listened to it also share my thoughts. I've measured several home audio systems (some really high-end systems) and they all show similar behaviour as the FR in this measurement.

Here's another plot on one of my headphones (AKG Q701) I believe sound tonal balanced (I do use some low end EQ to bring up the response below 80Hz):



Logically, one would assume a system in car to sound similar to the home audio system and the headphones if tuned similar, right? Well - car audio is the prime example of different practices, flat response simply sound bad. I dare to say - ANY system tuned flat (acquired with the methods explained in my other thread) will sound awful. Many have probably tried to tune their systems to flat and reached the same conclusions. So how would a balanced car audio system look like? More or less a downwards tilting response. This trend can be seen in every good sounding vehicle I've had the pleasure to both measure and listen at.

Once I conducted a little test, how accurately I use replicate the sound of the midrange (vocal tonal balance) of reference headphones (Sennheiser HD800) in my car. After 5 days with different listening sessions I came up with this curve;



What I was interested in what the important part of the vocal range, roughly 120-9000Hz. What's interesting is that while the headphones had a FR deviation of roughly 5dB +/- between 120-9000Hz, the car system required roughly a 15dB deviation (slope) to sound similar (full system, both sides playing).

This trend can be seen a variety of published articles. This is target response curve from Harman:



As seen, the downwards tilting slope is required to acquire tonal balance. IIRC, this curve is used in the JBL MS-8 together with binaural microphones. This acquisition method alters the tilt of the curve. Other well known projects of the community is the Magic Bus. His target response also show this trending (link below):

Measurements : Magic Bus

Here's another example, my personal high-score (tonal accuracy) on competitions I've been at came from using this curve;



All right. So if we establish that a downwards sloped curve is preferable, is there a perfect target response? I.e can you just copy someones curve and attain the same tonal balance as their system? This is a subject I've put a lot of research into and found that the answer is most often no... You can't copy a target response and expect the same tonality as another car using the same curve. The natural question would be; Why?

I have no definitive answers to this, but I believe the main reason is that different installs simply sound different. I'm talking about the speaker locations and what I'd like to call "reflection profile". A 2-way with speakers low in doors and tweeters in sails require a different target response than a 3-way with midrange high in doors or sails/pillars and tweeters nearby in pillars/sails. This is probably ONE of the reasons why auto EQ setups have varying success creating a good tonal balance. The other big reason being acquisition - How was the measurement acquired (method).

It's important to know that a measurement can look very different depending on how you acquire the data. I can measure the same system in several different ways and get different results every time. So that's why I established my own standard on how to do acquire the data, so others could replicate it and post data which was/is easier to interpret correctly (explained in the measurement thread). What this means is that if I setup my system using the same measurement method I can replicate the same tonality every time in the same install.

After all this rant, I can say what been said a hundred times over. Use the target curves as a baseline to get your system "nearly there". Or simply determine your tonal balance preference and build your own curve based on that. I did this and sort of interpolated the curve to make it flatter between the points I corrected for.

I can say that in every single case, a smooth transition sounds better than a sharp transition. This means that while the curve shouldn't be flat throughout all audible octaves, the transition between different sets of frequencies should be smooth when you're aiming for a target response.

------------------------------------------------------------
Another discussion I've come across on several occasions is whether to use pink noise or white noise together with an RTA. This is highly relevant, pink and white noise are different. See picture below:



What this picture is telling us is that white and pink noise got different energy composition "Spectral energy density" as it's referred to as in the picture. If you would tune for flat with white noise, it would display a falling slope if later measured using pink noise. If someone would be really obliged to listen to a "RTA flat system" white noise could be used as excitation signal. A system that measures flat with white noise will sound a lot more balanced than a system that measures flat with pink noise. This statement is highly valid within the realms of car audio since we want a falling response.

Here's a real world example. I was using the new APL DSP which I used a flat setting. Converted into pink noise spectrum we easily see the difference.



While I'm at it I might as well discuss correlated vs uncorrelated noise. It's also part of different acquisition methods. Correlated noise is noise recorded in mono. Uncorrelated is stereo based noise. You can use these for different things. Correlated can be used to center the image with T/A or EQ while uncorrelated often is used to measure the FR of the entire system with all speakers playing. They also can be used the approximate the FR of mono content in songs (often vocals) and stereo content. In a 100% time coherent system they shouldn't differ much.

I'll continue discussing this in the subgenre of this part next time.
 

·
Registered
Joined
·
693 Posts
Awesome, keep it up man!
 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #6
Part 2, Continuation on Target Responses...

I've gotten some questions on how 2-way and 3-way systems differ in terms of target responses. I should rather say "Install types" because it's not necessarily dependent on how many speakers you use, it's more about where they are located vs the listening position. A very common install type is factory locations with mids low in doors, often a 6-7" driver and tweeters high in doors/sails or pillars/dash.

The main issue with these types of installs is that it's hard to attain any proper center focus. The center console reflections is one big issue. Even though the center might sound diffuse, the stage is often large and perceived deep. I have not actually made a lot of measurements of cars with this type of setup. This measurement below is from a car that received fairly good rating on tonality during a competition a few years back.



This is mids+tweeters playing (full system). I didn't have any fullrange measurements of the car for some reason... Well, anyway, the transition from the lower midrange to the upper midrange is steeper than the other target response I posted earlier and yet it sounds balanced in this vehicle.

I believe the distance between the mid and tweeter is the big reason in this. If you aim for a flatter target, you will have the tweeters way too prominent and you might get that "in your face" staging and a lot of harshness if you're unlucky. That's my experience with it. Some of the best speakers in terms of "out-of-the-box performance" with little or no EQ had responses similar to this one;



If you imagine the tweeters a little bit attenuated with the crossover and the offaxis response of the woofer (with even more attenuation above 1,5kHz), we get a target response with a similar shape to the measured response above (somewhat at least) ^^
 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #7
Let's discuss acquisition methods a bit more...

There's a lot of discussion around whether to use Noise/RTA vs Tone Sweeps. Let's break it down and look at the pros and cons of both.

Sine Sweep - Measurements

This is common method of acquiring data. Used in RoomEQ (REW) and aRTA among others. A sine sweep is basically what it sounds like, you decide the frequency range it will play the sweep, the volume and the "length" which basically how long it it takes for the measurement to compete, this is also dependent on sampling rate of your soundcard - a higher sampling rate require a longer "length" of the measurement. A longer "length" provide better SNR (signal-to-noise ratio) than a short one. SNR can also be improved by using several sweeps (see picture below).

This is useful if you're measuring in noisy environments!



For the sake of convenience I will focus on using RoomEQ for now. Usually we are using 16-bit resolution with 44,1kHz or 48kHz sampling rate. We can only attain proper results that is half of this frequency, so for a 44,1k sampling we can get meaningful data up to 22,05kHz and 24kHz for a 48kHz sampling rate. A standard 16/44.1k resolution is adequate in most "non-technical" measurements. Don't sweat it.

Anyway, about the things you can attain by doing a sine-sweep:

*Magnitude and Phase Response = Frequency Response
*Harmonic Distortion Sweeps (2nd-10th order)
*Waterfall/CSD/Decay plots (displays linear distortion, "ringing in the time domain")
*Impulse/Step Response
*RT60 (Acoustical data, SPL vs Time decay in different frequency ranges)
*Group Delay

(We will discuss the uses for these later on)

[x] With a Sine-Sweep you are measuring ONE POINT in space. The time based data cannot be averaged and is only valid at this point in space. To get averaging we need to measure several points and sum them together, when we do this the averaged response will only retain its magnitude response, all other data will be lost. The power with sine swept measurements is the option to view all individual points separately and compare them to the averaged response for troubleshooting etc.

[x] Sine sweeps are very resilient to external noise, i.e background noise while measuring. This is also a powerful feature if you can't be in a quiet place.

[x] You need to do multiple sweeps to attain proper averaging, 3-8 points (depending on frequency range) is often enough if you not aiming for insane resolution.

That's the short version...


Noise / RTA - Measurements

This is another VERY common method. You use an excitation signal (most often pink noise) and play it on repeat through your system while recording it with an RTA. The RTA can be a physical device but most commonly nowadays it's just software. RoomEQ have this feature as well. So what can noise/RTA do?

*Magnitude Response

That's it. It only records SPL vs Frequency and nothing else. The RTA function itself can have various uses depending on the excitation signal used though. RTA/Noise is also more sensitive to external noise than sweeps.

The only really good thing about noise/RTA is that you don't require any smoothing in the software, you perform the averaging while measuring by moving the microphone around. 200-300 points in "infinite-averaging" can be viewed without any smoothing if you collected points in an adequately large area. This is a very accurate way to display the magnitude response. It's also easier to do than sweeps (imo).

We will discuss the advanced uses for these later on.
 

·
Registered
Joined
·
368 Posts
You can tell already this is going to be one of those threads that people get directed to for information. Thanks Han!
 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #14
How was that measurement done? What Signal did you use? Measured around head area?

More than 30dB between 100Hz and 10kHz seems far to much difference for me.
This wasn't my measurement actually. Since the guy who did it had no aux input I'm certain this is a full system with RTA/pink noise playing. He measures the same way I do it around the "headspace". I wouldn't put much emphasis on any of these plots - my point was that target responses can vary quite alot between install types/vehicles, that's environment and acquisition method plays a role in it (I'll come to this later).

If we do several measurements over a wide range of cars with the same install types tuned for good tonal balance we can get a mean value how the trending look (the shape of the curve). That would be a good starting point for a "general curve". Unfortunately my data is limited and have to rely on data made of other people as well. Having said that, I have tried to tune 2-ways (mids low in doors/tw high up) the same way as 3ways with midr in dash heights and the result was not pleasing, I believe the tweeters need attenuation (more than usual) using these types of install. How much and why? Not easy to say.

Sent from my Samsung Galaxy S5 using Tapatalk
 

·
Registered
Joined
·
634 Posts
I recently swapped my front stage for scans and this is my system as it stands right now using one of Hanatsu's curves as a baseline. Left side is Green. Right is purple. There's a couple of peaks and dips on the Right channel that I'm having trouble with it but overall it sounds pretty good.

 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #16
OK. This is hopefully the last 'main' post regarding target responses.

I'm sure you have seen this:



This is equal loudness contours (ELC). It's an estimation of what we perceive as equal volume over a span of frequencies. As seen by the picture, the plot differs as you change the volume. It basically says that at low volume, we are insensitive to low frequencies and that difference becomes less as we increase volume. How does this affect our target response?

Let's take a target response as an example;



Now, let's add the ELC curves on top of the target response;



What happened?

This is the same target response, more as it's perceived at different volumes! The ~SPL number mentioned in the picture is average midrange volume. If you tune your car for a competition you want to tune at a volume that's close to the volume to intend to play at during the competition. If you want to measure what target response you used as your own reference, make a note of what volume you measured at (in SPL @ x Frequency).

Perhaps something to think about...

-----------------------------------------

Ok. Next part...

Full system measurement or only one side + sub?

This is another ordinary question that shows up.

If you measure left side (or right) + sub individually you will get different results than if you would measure the entire system. I'm not talking about the midrange deviations that might occur, I'm talking about the low frequency shape.

In the region where the subwoofer meets the midwoofers, there will be three drivers playing the same range. If the drivers are playing in phase, then they will sum creating a larger boost in lower frequencies.

When posting a target response, it should be FULL SYSTEM. That's how you listen to it. Often we are measuring L/R independently and setting each side to match a target, be aware that you might get more lows than you asked for. A full system measurement should be performed in the end to double check the lower frequencies.

-----------------------------------------

Another little tip one smart guy told me was that if you setup your car in a small space (like a small garage) you will have more bass than if you tuned outside. The bass that "leaks" out into the room, will "leak" back in again. I've actually measured 1-2dB difference in my moderately sized garage. Might be neglectable, but you might want to take it into account when fine-tuning.
 

·
Registered
Joined
·
1,999 Posts
Another good reading!

Full system measurement or only one side + sub?

This is another ordinary question that shows up.

If you measure left side (or right) + sub individually you will get different results than if you would measure the entire system. I'm not talking about the midrange deviations that might occur, I'm talking about the low frequency shape.

In the region where the subwoofer meets the midwoofers, there will be three drivers playing the same range. If the drivers are playing in phase, then they will sum creating a larger boost in lower frequencies.
Oops I totally missed that.
It explains why I'm satisfied with the low end when playing, even if it looks a bit too flat on the side+sub measurements.


When posting a target response, it should be FULL SYSTEM. That's how you listen to it. Often we are measuring L/R independently and setting each side to match a target, be aware that you might get more lows than you asked for. A full system measurement should be performed in the end to double check the lower frequencies.
How do you measure the full system?
By averaging left&right in REW, or with a new measurement like 4 on left ear and 4 on right ear?

Also about the sub, measured alone I got variations between L&R, few db at 80Hz (and much more in higher frequency but not important).
I used few sweeps with mic on left ear only, right only, nose only, or both ears moving the mic.
 

·
Registered
Joined
·
4,939 Posts
Discussion Starter #19
How do you measure the full system?
By averaging left&right in REW, or with a new measurement like 4 on left ear and 4 on right ear?

Also about the sub, measured alone I got variations between L&R, few db at 80Hz (and much more in higher frequency but not important).
I used few sweeps with mic on left ear only, right only, nose only, or both ears moving the mic.
Subwoofer measurements require no averaging. One point is enough.

Full system is just what is sounds like, have all speakers on. Measure around the head with no bias on either side. I'll come to this later, different methods how to measure.
 

·
Registered
Joined
·
1,999 Posts
Subwoofer measurements require no averaging. One point is enough.
Well I have a little gap (left ear/right ear):

Not that I care for my tune, but was it just a wrong measure?


Full system is just what is sounds like, have all speakers on. Measure around the head with no bias on either side. I'll come to this later, different methods how to measure.
Good!
I have now a ton of measurements, and it helped a lot.
But I still get lost in the files, don't take enough notes on the context etc... I'd like to share this soon
 
1 - 20 of 29 Posts
Top